A new LogicSRC spec in the catalogs family. One heading, an identity block
(Kind, Web, Repo, Operator, License, Extends, Updated), one line, then eleven
layers: Languages, Runtimes, Interfaces (one ### per way in: web, api, cli,
tui, mcp, desktop, mobile, worker, extension, bot), Data, Services, Modules,
Tooling, Hosting, Auth, Conventions, Not. An item is one bullet: name,
version, an optional status word (trial, hold, leaving) and a role. Extends
inherits a parent file, sections replace, Conventions and Not accumulate.
Rule 8 is the reading rule for agents: use what is listed, prefer listed
over new, ask before adding a layer or a service, never add a Not, keep the
file true. Discovery at OpenStack.md in the repo, /.well-known/openstack.md,
rel=openstack, or a platform path. JSON is derived and never the source.
logicsrc.com serves its own at /.well-known/openstack.md: the route extracts
the worked example from docs/openstack.md, and a contract test holds the two
together and checks the file follows its own rules. rel=openstack in <head>
and in the Link header beside openprofile.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
The W3C AI Content Disclosure community group defines four attributes,
not one: ai-disclosure, ai-model, ai-provider and ai-prompt-url. Both
specs carried only the first, as `disclosure`. They now carry all four,
as `disclosure`, `ai_model`, `ai_provider` and `ai_prompt_url`: which
model, whose, and a page describing how it is used, optional and
meaningful only beside a disclosure other than none. `mixed`, the value
the group's meta tag form allows, is accepted at the site level in
OpenWebring and the page level in OpenWiki.
A ring host copies all of them into the member entry as the member said
them. A wiki carries them on the page and per revision, and a host
rendering a page puts the same four on it as the group's meta tag and
attributes. Both specs are 0.1.1; the wire version stays 0.1, since every
field is additive and absent is still unstated.
Claude-Session: https://claude.ai/code/session_01XYae2mH3khdwiXUVzcVMDw
Co-authored-by: Claude Fable 5.1 <noreply@anthropic.com>
OpenWiki writes down what the [[Page]] convention already is in practice,
so a wiki can be moved, mirrored, read by a directory and edited by an
agent without reading any engine's source. One Markdown file per page
with a small optional front matter (title, aliases, tags, summary,
author, made_by, disclosure, created, updated, license, lang, redirect);
wikilinks resolved by exact name, loose name, alias, path, then wanted;
one address with three representations (rendered, .md, .md?rev=) and a
history.json per page; a descriptor at /.well-known/openwiki.json with
pages_url, changes (Atom), repo, edit, editors and accepts; editing as
PUT of the same file with If-Match on the revision; discovery; what a
directory owes a wiki. Every revision says who made it: human, ai or
both, the same word OpenWebring uses.
A folder that follows it opens unchanged in Obsidian and Logseq, and any
such folder becomes an OpenWiki by adding the two files. The first host,
goviral.wiki, is being built on it.
Claude-Session: https://claude.ai/code/session_01XYae2mH3khdwiXUVzcVMDw
Co-authored-by: Claude Fable 5.1 <noreply@anthropic.com>
Manual publish of the report attached to the nixamp v0.24.2 release
(openstream-report-0.24.2.json/.md). The release workflow's
publish-report job skipped because LOGICSRC_PUBLISH_TOKEN is not set
on profullstack/nixamp. Built-in synthetic corpus, so the id carries
the -synthetic label per docs/openstream/reports/README.md.
Claude-Session: https://claude.ai/code/session_015uheT4Gn9BBKAnSZGcabyN
Co-authored-by: Claude Fable 5.1 <noreply@anthropic.com>
OpenL10n is the record of what a file says, in any language: one
transcript per media per language, addressed by the OpenFile id (or a
url: or live: identity for what has no bytes), lines as seconds into the
media, a translation kept beside the original and marked with what it
came from, complete or heard in pieces, made once on request with 202
progress, served as JSON, SRT, VTT or text.
OpenI18n is one file a service serves at /.well-known/openi18n.json
about the languages it speaks: which, which it can turn into which, how
to ask for one, and where texts are translated, with limits, models and
the pivot said out loud.
Both are what nixamp 0.24 and 0.25 serve today, written down so anything
that hears or translates can keep the same record.
Claude-Session: https://claude.ai/code/session_015KAKuRngFbQg3kET6RS5yN
Co-authored-by: Claude Fable 5.1 <noreply@anthropic.com>
The spec and its landing page now name the directory that reads every
OpenWebring host: one row per ring and one per member, with made_by as
the member said it and the status the host last verified, re-read hourly.
The first host line also names rssamplifier's curated Profullstack ring.
Claude-Session: https://claude.ai/code/session_01XYae2mH3khdwiXUVzcVMDw
Co-authored-by: Claude Fable 5.1 <noreply@anthropic.com>
* OpenFleet reference implementation: @logicsrc/openfleet 0.1.0 and logicsrc fleet
Ship what docs/openfleet.md describes. The new workspace package holds the
record (write once, never overwrite, 0600), the ledger (append-only JSON
Lines, merged across ledger*.jsonl by at), the ceiling rules (whole fleet
ceiling, narrowed swarm keys, a merge that never widens, refusals by key),
claiming and deriving exactly as the spec's "Claiming and deriving" and
rule 13, and fold(), which turns any $OPENFLEET_HOME plus the engine
rosters into the tree the landing page shows.
logicsrc fleet open|cap|tree|stop|log are the sysop's verbs, every one with
--json. open and cap exit 4 when OPENFLEET_MEMBER is set; stop exits 4
outside the caller's subtree, ends nested swarms first, goes through each
member's own engine (claude stop, moshcode herd kill, tmux kill-pane, a
signal for claude-p) and writes one swarm.end per swarm. tree reads claude
agents --json --all and ~/.moshcode/herd/sessions.json when it can, draws
recordless sessions as roster roots of the implicit fleet, and writes
member.end lost for a recorded member its engine no longer lists.
Claude Code takes part through hooks: logicsrc fleet hooks install merges
SessionStart, UserPromptSubmit, PreToolUse, Stop and SessionEnd into
~/.claude/settings.json without clobbering it, and logicsrc fleet hook
<Event> runs each one. SessionStart claims, derives or writes a root record
and hands the member its variables through CLAUDE_ENV_FILE; UserPromptSubmit
checks the ceiling with the permission mode the engine reports and writes
member.start, or refuses the first prompt with exit 2 and ceiling.refuse;
PreToolUse denies an edit outside piece.owns; Stop and SessionEnd write
member.end. A hand-started root takes the engine's reported approvals
before member.start, since the command line only guesses them. Hooks
never fail the engine: everything is caught and logged to hooks.log.
The spec and the landing page now say what ships, keep Status 0.1, and
record the two verified Claude Code limits: a background job dispatched from
claude agents gets no launcher environment, and OPENFLEET_* exported at
SessionStart reach the member's tools but not later hooks, so hooks key on
session_id through $OPENFLEET_HOME/sessions/<session_id>.json. PRD 0008
covers the work. CLI 0.2.1 -> 0.3.0; build and build:cli chains build the
package before the CLI; README and docs/cli.md list the group.
Tests: 95 in the package (record, ledger merge, every narrower case, the
worked example's claim and derive, the folded tree, hook install
idempotence, each hook handler including the exit-2 refusal and the
PreToolUse deny, every verb with fake deps) and 4 in the CLI.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01RZV4zJ2pDZLNN3kE5jFCmV
* OpenFleet fix round: rebuild the ceiling from the ledger, once-markers, rule 6 in tree, lost only for what a roster can hold
The review of the reference implementation against moshcode found the two
readers disagreeing on the same files. This round applies the shared
rulings so both sides read a ledger the same way.
Ceiling (R-A, R-B, R-C, R1, R6, R10, R15, R17): memberCeiling rebuilds the
effective ceiling from the ledger on every read. The latest fleet-target
fleet.cap (else fleet.open, else the implicit fleet's) replaces the copy in
a record, so a sysop's widening cap reaches running members; then each
swarm.spawn narrowing down the path, then swarm caps last. In the implicit
fleet a parentless record's own approvals enters at the root; a ceiling a
writer left without the key is never read as native, and startMember fills
it with the engine's word while the record is unclaimed. A fleet.open or
cap with no hosts means the host it was written on (R23).
Once-markers (R-G, R28): member.start, member.end and swarm.end each take
an exclusive create under fleets/<fleet>/marks/<event>.<id> before the
append; a lost end takes <id>.lost so a real end can still supersede it.
The hooks let a real end follow a lost line (R9).
tree (R-F, R20): run by the sysop it enforces rule 6, stopping a member
past its effective until with state timeout and the members of a swarm or
fleet at its budget with state budget, then writes swarm.end for each swarm
touched once it is complete. An agent's tree stops nothing. lost is written
only for a member its engine's roster can hold: a claude-code background job
(8-hex member or session) or a moshcode pane, never an interactive session
claude agents does not list (R-E, R3, R14). A nested swarm is drawn under
the member that spawned it and its row shows the effective ceiling (R25).
stop and cap (R-D, R-H, R22, R27): swarm.end is written only once every
member and every nested swarm has an end line that counts; an engine that
will not end a member leaves it without an end line and the verb exits
non-zero. claude stop takes the job id: the member of a background job,
else the first eight characters of a session UUID; an interactive session
with no job id cannot be stopped and the tool says so. cap on a swarm
refuses a key that would widen. A derived claude-code job is named by its
job id and carries no pid.
Also: R-I (endMember ends only the engine-minted swarm of one), R35 (a
derived record's guessed approvals corrected at UserPromptSubmit), R32
(the UserPromptSubmit hook passes only exit 2 through), R31 (package
README), R36 (rule 13 says the launcher test is unimplemented in 0.1),
docs and PRD 0008 updated for lost, rule 6 and the markers. 113 openfleet
tests, 93 CLI tests, contract green.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01RZV4zJ2pDZLNN3kE5jFCmV
* openfleet hooks: no member.end for a member that never started
A first prompt refused by the ceiling still lets the session wind down through Stop and SessionEnd; those handlers now write nothing when the ledger holds no member.start for the member, so a refused member is never drawn as done.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01RZV4zJ2pDZLNN3kE5jFCmV
* logicsrc-mcp test: the next free PRD id is 0009 now that PRD 0008 exists
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01RZV4zJ2pDZLNN3kE5jFCmV
---------
Co-authored-by: Claude Fable 5.1 <noreply@anthropic.com>
PR 177 shipped a second OpenFleet under the same slug: a published listing of OpenAgent profiles and ipfile swarms with CoinPay rental offers. Anthony ruled that OpenFleet means the human-controlled fleet and the record of who spawned whom, so the rental contract is renamed OpenRental (slug openrental, catalogs family, group noun listing): docs/openrental.md, logicsrc-openrental.schema.json with type logicsrc.openrental and scope kind listing, fixtures/openrental, createOpenRental and OpenRental* types in the SDK, the openrental validator kind. Minor bumps because an export moved: @logicsrc/schemas 0.2.0, @logicsrc/validators 0.2.0, @logicsrc/sdk 0.2.0. The discovery contract test now covers openfleet, openrental and openwall, one per family, and accepts a landing-page link in llms.txt.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01RZV4zJ2pDZLNN3kE5jFCmV
One file a ring serves about its members, one file a member may serve
about itself, plain links between them, and a made_by declaration on
every member (human, ai, both) so a reader can follow the human web or
the machine web on purpose. Six hop rules that accept every addressing
shape rings already use (?from=, ?host=, ?via=, ?url=, a slug in the
path, the bare Referer), so a member of any existing ring joins with no
change; verification the IndieWeb way (mark inactive, never delete); an
OPML twin of every ring; no script, no tracking, no central registry.
Name: OpenRing is Drew DeVault's tool in this exact niche and every
openring domain is taken; OpenWebring has only a dead 2020 predecessor
and free domains. Registered in the catalogs family.
First host: rssamplifier.com/ring (one ring per topic), in flight.
Claude-Session: https://claude.ai/code/session_01XYae2mH3khdwiXUVzcVMDw
Co-authored-by: Claude Fable 5.1 <noreply@anthropic.com>
A new spec at /openfleet and /docs/openfleet replacing the AgentSwarm placeholder. A fleet is every agent session one human answers for; a swarm is the members one spawner starts inside it for one task. One record per session (claimed by the first session that writes its member.start, derived when inherited), one append-only ledger per fleet per host with eight events, five sysop verbs, fifteen rules, and what Claude Code and moshcode would each add. Written from job 172ffd83, which had to reconstruct its own parentage by hand.
openswarm keeps its slug (it is the peer-to-peer media family) and gains a one-line pointer. /agent-swarm redirects permanently to /openfleet; the registry entry, home band, catch-all route and scroll hook for the placeholder are replaced or removed. Contract test covers the redirect.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01RZV4zJ2pDZLNN3kE5jFCmV
* feat: add OpenFleet membership and CoinPay rental contracts
* Propose OpenWall broadcasts and direct messaging contracts
* release: prepare OpenFleet and OpenWall public contracts
* fix: use tested npm for compatible CLI installs
Every consumer of a link reads a page's card tags its own way, caches the
first reading for days, and publishes nothing. OpenSite writes the reading
down: a record per URL (title, description, image, kind, canonical, author,
feeds, the card tags verbatim, JSON-LD as parsed), a descriptor a site
serves at /.well-known/opensite.json, the order a reader takes each field
from a page, the four calls an index offers, and a table of what each
consumer reads and caches.
docs/opensite.md, one entry in the specs registry (catalogs family), and a
landing page at /opensite. The first index is nichedb.dev/c/sites; the
first publisher is nixamp's share links.
Claude-Session: https://claude.ai/code/session_01MwAoNvWzezmBHeT7oDHo3C
Co-authored-by: Claude Fable 5.1 <noreply@anthropic.com>
Subscribing takes one click and cancelling takes a support ticket, and no
reader can check that the two are as easy as each other because the way
out is not written down anywhere a reader can fetch. The service already
has a plan table, a cancel endpoint and a refund rule. OpenSaaS is those
at /.well-known/opensaas.json: plans, and eight actions (subscribe,
cancel, pause, resume, change_plan, unsubscribe, export, delete), each
as the page a person opens and the endpoint an agent calls with an
OpenAccess scope, with steps and confirm as the exit measure a directory
shows beside the entrance. Doc, landing page, registry entry under
catalogs. nichedb.dev is named as the first directory and the first
service.
Claude-Session: https://claude.ai/code/session_01S7yeJUHGxA4P5N74xnsRPQ
Co-authored-by: Claude Fable 5.1 <noreply@anthropic.com>
The SimpleFIN door written down for anything a bridge holds: a person gets
a single-use setup token from the bridge, pastes it into an app, the app
claims it once for an access URL and a bearer the bridge can revoke. No
client registration, no redirect, no key for the app to keep, which is what
a browser extension or a script needs. Bearer instead of SimpleFIN's Basic
credentials in the URL, because a browser's fetch refuses those. Eight
rules, the social profile (accounts, analyze, write, suggest, activity,
posts only when declared) and the finance profile (SimpleFIN, unchanged).
First bridge: mynaposter.com (/connect). First app: DefPromo. Registered
as one entry in the specs registry under Access and credentials, beside
OpenAccess, which is the registered door with the same scope vocabulary.
Claude-Session: https://claude.ai/code/session_01XYae2mH3khdwiXUVzcVMDw
Co-authored-by: Claude Fable 5.1 <noreply@anthropic.com>
A publisher and a disk each say which CLI drives them, in the same
`developer` shape OpenServer 0.2 defines, so a marketplace shows one
install line beside every listing without reading a docs page per host.
fi1zes.com is the name chosen for the OpenFile marketplace on 2026-09-13,
not yet registered; bittorrented.com stays the reference reader until then.
Claude-Session: https://claude.ai/code/session_01Khk1C6Ese6xjdHAWLVstca
Co-authored-by: Claude Fable 5.1 <noreply@anthropic.com>
How a machine should sound when it speaks for you, and the order a reader
chooses in: Voice, then Gender, then Pronouns, never a name or a photo.
Gender means the same in the identity block and under Match. The first
reader is nixamp's party line, which reads trollbox lines to the people
on the phone in a room in the author's voice (nixamp 0.23.5).
Claude-Session: https://claude.ai/code/session_01GxYGCCvvuhVAkU1W2iJKTV
Co-authored-by: Claude Fable 5.1 <noreply@anthropic.com>
A directory that lists a host's plans is asked the next question at once:
how do I drive this thing from a terminal? Every provider answers it on a
page of its own, in its own words, and a reader that wants the install line
has to find and read that page for each host. nichedb.dev's hosting
collection just did exactly that for 46 providers, copying the commands
off each vendor's guide, and the exercise is the argument for putting the
facts in the descriptor.
`provider.developer` names the official CLI, its install commands keyed by
package manager and copied as the guide prints them, the install guide and
source, the API docs, the Terraform provider and the GitHub organisation. A
provider with no CLI says `"cli": null`, which is a fact, while a missing
block means unknown. Commands are copied, never composed: a reader that
invents `brew install <name>` sends a buyer to a formula that may not exist.
c0mpute.md gains the same block for the compute market, with the one
install line c0mpute.com prints, and the landing page says what 0.2 adds.
Claude-Session: https://claude.ai/code/session_01Khk1C6Ese6xjdHAWLVstca
Co-authored-by: Claude Fable 5.1 <noreply@anthropic.com>
* docs: OpenBroadcast and OpenGuest, the broadcaster-and-guest framework as OpenProfile.md sections
Anthony: "broadcasters and guests is the usual framework for live audio
shows, radio, podcasts" and OpenProfile.md should carry it so a platform
(anyfans) can match hosts with guests from two files rather than two
forms. OpenExpert folds into OpenGuest: an expert is a guest with
Expertise and Credentials.
OpenBroadcast is the `## Broadcast` section: Show, Kind, Format, Live,
Cadence, Length, Language, Audience (host's own unit), Feed, Topics,
Seeking, Not, Slots, Remote, Book, and Pays / Charges (unstated by
default, because pay-to-play is the thing a guest is most often not
told). OpenGuest is the `## Guest` section: Available, Expertise,
Credentials, Pitch, Formats, Live, Languages, Availability, Lead time,
Remote, Rate, Pays, Appeared on, Press, Book, Not. Matching scores
Topics/Seeking against Expertise/Topics, Slots against Availability,
Pays/Charges against Rate/Pays; both Not keys are absolute; a platform
never fills a key the person did not write. Both landing pages share
profile-section-page.tsx. OpenProfile.md names the two sections in rule
4 and in Related standards. Registered in the four places.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_014cmNRtR2vL1p89dbVQ7FZJ
* ci: trigger workflows
---------
Co-authored-by: Claude Fable 5.1 <noreply@anthropic.com>
* docs: OpenThreat, one file a security tool serves about what it found in the open
Twelve rules that degrade and two that do not: a subject is public or it
is not in the file (no private repos, no customer servers, no paid users'
scans), and a secret is never located while it is open (rule, severity,
subject, status only; no location, message or excerpt). Four kinds:
finding, attack, indicator, advisory. Status open, fixed, mitigated,
blocked, withdrawn; a withdrawn threat stays a while so directories
retract it. Announcing is on by default with a one-switch opt-out in the
tool's own settings. Discovery at /.well-known/openthreat.json,
rel="openthreat", or a handed URL; origin is the verification. Mapped
against SARIF, STIX 2.1 and CSAF rather than replacing them.
First reporter: threatcrush.com/discovery (its own PR). First directory:
nichedb.dev/c/threats (its own PR). Registered in DOC_SLUGS, NAV,
STATIC_ROUTES and llms.txt.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_014cmNRtR2vL1p89dbVQ7FZJ
* ci: trigger workflows
---------
Co-authored-by: Claude Fable 5.1 <noreply@anthropic.com>
The descriptor at /.well-known/openaffiliate.json (programs with pays,
link, window, attribution, hold_days, payout, approval, self), the four
calls (join with an OpenProfile.md, link with ?oa=code, read the ledger,
get paid to your own address), webhooks, discovery, what a directory
owes a merchant, and what is deliberately absent: no network in the
money, no tracking host, no application form, no exclusivity, no
impression payments. Landing page at /openaffiliate, registered in
DOC_SLUGS, NAV, sitemap and llms.txt. Reference implementation is
crawlproof.com/affiliate.
Claude-Session: https://claude.ai/code/session_01CDEiDss9RWYibtmxSk5Gr2
Co-authored-by: Claude Fable 5.1 <noreply@anthropic.com>
Anthony: every top-level nichedb.dev niche may need its own open<niche>
spec so the serve-your-own-file pattern scales across industries. These
are the two he named first.
OpenCoupon: one JSON file a merchant serves at
/.well-known/opencoupon.json about what is on offer right now: every
code, sale and shipping threshold with kind (percent, amount, shipping,
bogo, gift, other), value, scope, min_order, dates, status, per-customer
and region limits. Expired coupons stay in the file so a directory
learns a code died from the one party that knows. No affiliate links,
no redemption, no votes. First reader: nichedb.dev/c/deals.
OpenRecipe.md: one Markdown file that is a recipe, in the OpenProfile.md
and OpenResume.md style: a summary block (Serves, Prep, Cook, Cuisine,
Course, Diet, Author, Source, Image), a description line, Ingredients
and Steps as written, Notes, Nutrition per serving. Served next to the
page, linked with rel="openrecipe", or indexed at
/.well-known/openrecipe.md. A one-way mapping to schema.org/Recipe:
the JSON-LD is generated from the Markdown, never the reverse.
Both registered in DOC_SLUGS, NAV, STATIC_ROUTES and llms.txt.
Claude-Session: https://claude.ai/code/session_014cmNRtR2vL1p89dbVQ7FZJ
Co-authored-by: Claude Fable 5.1 <noreply@anthropic.com>
OpenSwarm says how a file lives on a swarm and how a seeder is paid to hold
it, and nothing in it gives a person with a browser, a search engine or a
directory a URL to start from. A manifest is on the DHT under a key, the
catalogue is a feed found through the DHT, the seed market is one-sided:
offers are listed and seeders poll them. A requester who wants a box in
Germany with two terabytes free and a year of clean proofs has no file to
read, and a seeder with those things has no file to serve.
OpenFile is /.well-known/openfile.json on a publisher's origin: each file
by its plaintext content hash (the ipfile plainRoot, so the id a reader
gets is the root the decrypted file verifies against), every way to fetch
it (an ipfile magnet, a webseed, plain HTTP by range, HLS for a player
with nothing installed), the pay2seed attestation and README, the price
as an x402 offer URL, and a holders list of who has the bytes now with
the age of each seeder's last proof. Encryption is ipfile unless the
publisher says none as an explicit act.
OpenDisk is /.well-known/opendisk.json on a machine that rents disk: free
GiB, price per GiB-month in the unit pay2seed already prices in, the
operator's accept policy stated up front so nobody posts an offer the disk
would never take, proof cadence, the seeder key and the hubs it takes
leases at, and a record block whose source is the hub's own seeder page,
because a marketplace reads standing from the hub and never from the
file. d1sks.com is the reference marketplace. A disk is also an OpenServer
offer of kind storage, and the mapping is a table, so the nichedb hosting
collection lists every disk without a second parser.
Neither restates a record that already has a name: the swarm is ipfile,
the consent is pay2seed, the leases and proofs are paid2seed, the payee
and the pass are ippay, the feed is ipdb. Both are registered in the four
places a LogicSRC spec needs and added to the OpenSwarm family table.
OpenFile has no product domain yet and says so.
Claude-Session: https://claude.ai/code/session_01Khk1C6Ese6xjdHAWLVstca
Co-authored-by: Claude Fable 5.1 <noreply@anthropic.com>
Four resource specifications under OpenServer, one per thing that is
negotiable when a server is bought. Each is the block of an OpenServer
offer (compute, memory, gpu, network) written down on its own, with the
units OpenServer 0.1 already uses (vcpu, cores, ram_mb, vram_mb,
bandwidth_mbps, transfer_gb, ipv4, ipv6) and one new shape shared by all
four: `range`, the field a buyer can dial at checkout, its bounds, the
step and what a step costs on top of the base price.
- OpenCPU: threads against cores, the processor by its vendor name,
dedicated, shared or burstable allocation.
- OpenMemory: mebibytes, DDR generation, ECC as three states, reserved,
balloonable or shared; wins over compute.ram_mb when both are present.
- OpenGPU: the card by its vendor name, count and VRAM per device,
interconnect, passthrough, MIG, vGPU or shared access.
- OpenBandwidth: port, four meters (transfer, unmetered, percentile,
flat), overage, IPv4 and IPv6 addresses as a priced resource.
A provider that sells only one resource lists it as an OpenServer offer
and may serve the same document at /.well-known/<slug>.json. Landing
pages share one component (resource-spec-page.tsx). Registered in
DOC_SLUGS, NAV, STATIC_ROUTES and llms.txt. OpenServer, OpenFile and
OpenDisk arrive in sibling PRs.
Claude-Session: https://claude.ai/code/session_014cmNRtR2vL1p89dbVQ7FZJ
Co-authored-by: Claude Fable 5.1 <noreply@anthropic.com>
Rule 9, `## Match` (Dating, Matching, Partner and Looking for normalise to
it): the keys a matching platform needs, about you (Born, Gender,
Orientation, Status, Monogamy, Height, Body, Children, Wants children,
Smoking, Drinking, Cannabis, Drugs, Religion, Politics, Ethnicity,
Education, Work, Diet, Pets, Exercise, Zodiac) and about who you seek
(Seeking, For, Ages, Distance, Not). Values are kept as written and
matched loosely like Topics; unknown keys are kept; absence is unstated.
Two rules that do not degrade: Born is the one key a matching platform
must have, and a computed age under 18 keeps the profile out of any
matching context; and the section is public by nature, so a platform
stores only what the person confirmed with it and drops it when the
file does. A `## Photos` section carries image URLs, first is the lead.
"No inference" joins the deliberately-absent list.
Claude-Session: https://claude.ai/code/session_014cmNRtR2vL1p89dbVQ7FZJ
Co-authored-by: Claude Fable 5.1 <noreply@anthropic.com>
* OpenServer 0.1: one file a hosting provider serves about what it sells
A new LogicSRC spec at /openserver and /docs/openserver. A provider puts
the table its order form already reads at /.well-known/openserver.json:
every offer with a kind, four axes, specs in fixed units, one price,
location and stock. A directory reads the provider instead of scraping
an aggregator whose terms forbid it, and the provider stays the author
of its own catalog. Only provider.name and each offer's name are
required; every other rule degrades.
Fifteen kinds cover what Anthony listed and the rest of the market:
cloud, vps, dedicated, bare-metal, colocation, on-prem, shared, managed,
paas, serverless, storage, gpu, edge, p2p and hybrid. Premises,
management, tenancy and model are their own keys rather than inferred
from the kind, because a managed VPS and an unmanaged one are the same
kind and different offers. A peer-to-peer market publishes one
descriptor whose offers are its current asks, with the operator pointing
at the market and not the peer; c0mpute is the compute case, OpenDisk
the storage case, OpenSwarm the settlement layer under both.
The first reader is nichedb.dev's hosting collection, being built
alongside this. findhost.app is named as the curated sibling.
Registered in DOC_SLUGS, NAV, STATIC_ROUTES and llms.txt, one line each.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Khk1C6Ese6xjdHAWLVstca
* OpenServer: name the resource blocks an offer may carry
OpenCPU, OpenMemory, OpenGPU and OpenBandwidth are being written as the
blocks that nest inside an offer's compute, compute.gpu and network, and
that stand alone as offers. Related standards now says so, with links at
/docs/<slug> where those specs will land. No subdomain is named anywhere
in this spec: every LogicSRC spec lives on logicsrc.com only.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Khk1C6Ese6xjdHAWLVstca
---------
Co-authored-by: Claude Fable 5.1 <noreply@anthropic.com>
A new LogicSRC spec at /openaccess and /docs/openaccess. A person, an
agent or an organisation keeps one account at a hub; every app keeps its
own users and links each to that account once. Grants delegate narrower
to agents, and a subscription bought in one app is honoured by every app
that honours the product. The app descriptor, the hub metadata, the four
flows, the token, signed webhooks and the hub's own doors.
Registered in DOC_SLUGS, NAV, STATIC_ROUTES and llms.txt.
Reference implementation: github.com/logicsrc/openaccess, hub at
openaccess.logicsrc.com.
Claude-Session: https://claude.ai/code/session_01SWRffW4ifQPUrGXJtgYWMd
Co-authored-by: Claude Fable 5.1 <noreply@anthropic.com>
OpenContext, OpenCreds, OpenOntology and OpenPRD were already registered
under the umbrella. Three were not:
- `logicsrc openmcp …`: OpenMCP lives in its own repository, so it is a
dependency (@logicsrc/openmcp ^0.3.1, which exports ./cli for this) and
every argument goes untouched to the same main the standalone binary runs.
Imported on first use, because the catalog is node:sqlite (Node 24) while
the rest of the CLI runs on 18; below the floor that one word says so and
offers the standalone installer, which brings its own Node.
- `logicsrc openspec <anything else>`: import, export and change stay ours;
any other word (init, list, validate, archive, show) runs OpenSpec.dev's
own CLI (@fission-ai/openspec) as the group's default subcommand, flags
intact. One command for a repo in compatibility mode, and the OpenSpec.dev
half is upstream itself rather than a copy that would drift.
- `logicsrc mcp`: the LogicSRC MCP server (@profullstack/logicsrc-mcp) over
stdio, spawned as a child because it owns the process's stdio. build:cli
now builds that workspace.
README gets a table of every word and the standalone name it mirrors; the
OpenMCP spec page and doc mention the umbrella form.
Claude-Session: https://claude.ai/code/session_01Qh2dieNyPZ4Hx5g3XNJ1Eo
Co-authored-by: Claude Fable 5.1 <noreply@anthropic.com>
logicsrc.com/.well-known/openprofile.md was a 404 while every relay in the
OpenMCP catalog, Obscura included, named it as its operator. The site that
publishes the spec now serves its own file: Kind organization, accounts
(GitHub, blog, the catalog, the Obscura relay so the trust chain closes both
ways), topics, projects and contact. The root layout carries
rel="openprofile" and every response carries the same relation as a Link
header, per the spec's discovery rules. skill.md and llms.txt point at it.
The OpenMCP spec page and docs/openmcp.md now lead with
curl -fsSL https://openmcp.logicsrc.com/install.sh | sh instead of npx, and
the descriptor examples name logicsrc.com's real profile rather than one
profullstack.com never served.
Claude-Session: https://claude.ai/code/session_01Qh2dieNyPZ4Hx5g3XNJ1Eo
Co-authored-by: Claude Fable 5.1 <noreply@anthropic.com>
did:key, did:web and AT Protocol did:plc, kept as written. A DID from a
service that also vouches for agents (CoinPay issues one per account and
lets a person's stand behind an agent's) is how Operator becomes checkable
rather than stated. myna did attaches one.
Claude-Session: https://claude.ai/code/session_01FMT2v1YxmgcDuionrfT719
Co-authored-by: Claude Fable 5.1 <noreply@anthropic.com>
A relay serves /.well-known/openmcp.json; a catalog probes it (the
descriptor from the relay's own origin, then initialize and tools/list)
and lists only what it found; a client reaches every relay through the
catalog's REST, its own MCP endpoint, or signed webhooks. Landing page at
/openmcp, the document at /docs/openmcp, registered in the same four
places as the other specs. Reference implementation at
github.com/logicsrc/openmcp.
Claude-Session: https://claude.ai/code/session_01FMT2v1YxmgcDuionrfT719
Co-authored-by: Claude Fable 5.1 <noreply@anthropic.com>
A new LogicSRC spec at /openprofile with the document at /docs/openprofile.
Eight degrading rules (name, identity block, headline, sections, accounts,
topics, reshare terms, operator), three discovery locations
(/.well-known/openprofile.md, rel="openprofile", a platform path) and
verification by linking back. Registered in DOC_SLUGS, NAV, STATIC_ROUTES
and llms.txt, the same four places as ASDLC.
myna writes one from its accounts and publishes the Reshare section to the
myna reshare network; agenticjobs serves one per public candidate.
Claude-Session: https://claude.ai/code/session_01FMT2v1YxmgcDuionrfT719
Co-authored-by: Claude Fable 5.1 <noreply@anthropic.com>
Adds a reports section to the OpenStream spec so its claims rest on a
reproducible measurement rather than an assertion. Each report is a run of
the envelope over a defined corpus on real hardware: proof that
decompression restores every byte, that an incompressible input costs only
the framing overhead, that a compressible one saves what it claims against
the complete wire size, and how long each codec takes.
- docs/openstream/reports/ holds a machine-readable <id>.json (canonical,
with a versioned schema) and a rendered <id>.md per report, plus a README
on the shape and on submitting one. The seed report is nixamp 0.17.1 over
the synthetic corpus, labelled synthetic so no one reads a padded-fixture
number as production.
- The site renders them at /docs/openstream/reports (index) and
/docs/openstream/reports/<id> (one report), under the dynamic /docs/[slug]
tree so the reports routes never shadow a spec's own doc page. A small
lib/reports.ts reads the JSON at build time; REPORTED_SPECS keeps the
route surface explicit. sitemap includes the index and every report.
- The spec doc gains a Benchmark reports section linking there, and repeats
the honest caveats: OpenStream frames Zstandard and gzip rather than being
a new algorithm, synthetic padding flatters a codec, an efficient real
feed saves little, and round-trip exactness is the one pass/fail.
The report format is produced by `nixamp compression benchmark` (in the
nixamp repo); a release runs it and commits the two files here.
Claude-Session: https://claude.ai/code/session_01MxNif5tsYq4LczgG7aE8Jp
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
OpenStream is the wire format nixamp uses to relay a channel or a file
between two servers: a framed stream of blocks, each compressed with an
established codec or stored verbatim, each carrying the length and SHA-256
of the bytes it stands for, ending in a marker that says the stream
finished rather than dropped. It is a framing envelope, not a compression
algorithm, and it is deliberately product-neutral: nixamp is the reference
implementation, the format carries any byte stream.
The doc gives the byte layout (16-byte stream header, 48-byte frames, both
big-endian), the mode set, the validation order, cross-language test
vectors, the negotiation, the source/channel boundary, recovery semantics,
and a conformance checklist. Published at /docs/openstream via the same
DOC_SLUGS path as every other spec; no README change, matching how
OpenJob/OpenResume (#148) landed.
Claude-Session: https://claude.ai/code/session_01MxNif5tsYq4LczgG7aE8Jp
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
Two conventions for the hiring end of the agentic stack, published here beside
the other Open* specs.
OpenResume.md says a resume is a Markdown file: a document a person can read,
diff and keep, and one an agent can write without being taught a schema first.
Six conventions, every one of which degrades rather than fails, because a
resume that does not parse still has to be a usable resume.
OpenJob extends schema.org JobPosting with the three things it has no
vocabulary for: whether the employer accepts applications written with an agent
(stated, rather than discovered by silent rejection), the application form as
data so applying does not require rendering a page, and a description in
Markdown.
Both are implemented by profullstack/agenticjobs, and neither requires it.
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
OpenPRD 0.2 fixed eight body sections, none of which asked what the thing is
built on or how it earns. The stack got chosen in the first implementation PR
instead of at review, and a PRD could be filled out completely without anyone
writing down who pays. PRD 0006 had already grown a hand-rolled
`## Business model` section, which is the gap showing.
0.3 adds two required sections between `UX Notes` and `Success Metrics`:
- Tech Stack — languages, frameworks, datastores, third-party services, and
anything the work must not depend on. It makes the requirements costable.
- Monetization — the revenue model: who pays, for what, how much, and when.
`_None._` stays a valid answer, but it now has to be said out loud.
Adding required sections would normally invalidate every document already
written, so a document is now held to the section list its own `openprd:` key
fixes. A 0.2 document keeps conforming with eight sections, forever; a 0.3
document needs ten. Adoption is per document, and `logicsrc prd validate
--expect-version 0.3` (new flag, wiring up the validator option that already
existed) reports the stragglers as OP-L-VERSION.
The front-matter schema is untouched — both additions are body sections.
Conformance bundle proves both directions: invalid/missing-monetization.md
fails with OP-C-SECTION-MISSING, and valid/legacy-0-2.md passes unedited.
This repo's own PRDs 0001-0006 stay at 0.2 as standing evidence that the
compatibility rule holds. PRD 0007 records the decision at 0.3.
Claude-Session: https://claude.ai/code/session_017XRNNm6pK6nPi7rJ6bJNHu
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
ASDLC 0.1 describes how software gets built when most of the work is done
by agents running in parallel and CI/CD is the only gate. It is a
description of a practice already in production, not a proposal.
The traditional SDLC assumes the scarce resource is engineering time, so
it spends process on deciding whether each change is worth building. When
agents write the code, engineering time stops being scarce and two other
things become scarce: human attention, and trunk stability.
Nine phases: frame, fan out, gate locally, merge, release, verify live,
correct, ratchet, promote. Correct returns to fan out, so the loop is the
lifecycle.
The load-bearing phase is the ratchet. Testing in production is only
defensible if production failures are one-time events, so every escape
becomes a permanent automated check before the incident is closed, and
that check has to be confirmed to fail when the bug is reintroduced. A fix
without a ratchet is how the same class of bug ships three times.
Four conformance levels, of which only level 3 requires evidence rather
than intent. The worked example is DiskPush on 2026-09-06: eight agent
worktrees on one checkout, four releases between 08:53 and 14:56 UTC, and
a three-release desktop bug whose first layer no local harness could have
caught, because a static server resolves absolute paths correctly by
construction and the bug only existed under file://.
Published at /asdlc with the spec at /docs/asdlc, listed in the nav,
sitemap and llms.txt.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012Q2bt449mEJSHoEZzaemCn
* Add pay2seed to the OpenSwarm family: consent at upload and a paid seed market
OpenSwarm pays a seeder per verified piece served, and nothing pays anyone
to stay. An archive, a backup, a dataset waiting for its buyer or a
podcast's back catalogue earns nothing the month nobody downloads it, so
it dies the way every swarm always has. And nothing in BitTorrent says who
put a swarm there or whether they were allowed to, which is why a seeder
is presumed to be doing something wrong.
pay2seed is the member document for both halves. An attestation, signed
at upload with a fixed basis (own, licensed, open-license, public-domain,
personal) and a notice endpoint, is what a hub requires before it will
list anything; public claims get a claim window and a standing, and a
notice voids them. An offer escrows a budget at an ippay hub for a swarm,
public or private, to be held by M seeders for N days at a price per
GiB-month, bought over x402 exactly as a pass is. Seeders take leases,
prove each period by storage challenge or by a probe over the ordinary
wire, and are paid through the payee they already have. Public feeds ride
on ipdb; ipfile.pin on c0mpute is the same offer on the auction.
Also: the family table, stack diagram and registry rows; the ip seed
command group; PRD 0006; the protocol row on /openswarm.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01SKAohrRkqLKVQL2cGCAkR5
* Split pay2seed into client and server halves, and add pay2stream and paid2stream
The rule is now in the names. pay2* is the client protocol: the side that
pays, over HTTPS, and plays. paid2* is the server protocol: the BitTorrent
side that earns. One hub implements both halves of a pair; a requester or
viewer implements only pay2*; a seeder, relay or gateway only paid2*.
pay2seed keeps consent, offers, the requester's market and notices.
paid2seed takes leases, storage challenges and probes over the wire,
GiB-month accrual and receipts, the seeder client, and the ipfile.pin
mapping.
pay2stream and paid2stream do the same for a live channel over iplive.
A broadcaster attests the channel (with the two rules that separate a
licensed rebroadcast from a stolen feed), buys relays by the hour, and
publishes listings; viewers buy tickets. Relays take leases and are
proven present by a verifier that pulls segments as a peer; a gateway is
a relay that also serves standard HLS, clear or sealed, with the M3U and
XMLTV pair every IPTV app asks for, so VLC, TiviMate, Kodi and a
television play a paid swarm with nothing installed. Ace Stream showed
BitTorrent can carry live TV to millions; this is that with consent,
payment and an open spec.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01SKAohrRkqLKVQL2cGCAkR5
* pay2seed: encrypted by default, access as the product, a README in every swarm
The client encrypts by default and the hub never does: what the hub
manages is who may decrypt. A team is a named set of member keys with a
scope over the owner's swarms; the hub, as keeper, issues grants to
members when the owner is offline, with invitations that expire, roles,
an audit trail, and re-encryption on removal so the next version is
closed to whoever left. A few seats are free; above that the hub charges
per seat and per organisation, settled through the same pay plugins as
everything else. Seeding is priced at disk; access is where a hub earns,
and both sides earn: seeders rent disk, requesters sell access.
Public is not a fallback. Encryption off is an explicit act, and a
public swarm is attested, listed, kept and rendered exactly as a private
one is; the only difference is who can read it.
Every swarm on the market carries a README.md at its root, no
exceptions, and the attestation carries its Markdown and the hash of
the copy inside the swarm, so the hub renders it as the swarm's page
without a key. Relative links resolve into the swarm and are gated the
way the files are.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01SKAohrRkqLKVQL2cGCAkR5
* pay2seed: 1 percent, ads on the free tier, and agents as sellers
Three things the specs did not say. The reference hub takes 1 percent of
any payment that crosses it, charged to whoever is paying and never
deducted from a seeder or a relay, so a quoted price is what the
publisher gets and a promised floor is what the seeder is paid.
Public swarms are free to fetch and free to list, and an advertisement on
the swarm README page is what pays for that. The ad is on the hub page
and nowhere else: never inside a swarm, never injected into a file, a
segment or a playlist, and never in the catalogue or the market API. A
requester who wants no ad buys a seat instead. A free-to-watch channel
works the same way.
And a requester is a key, not a person. An agent can attest what it made,
price access, sell tickets, take payment through its own payee and spend
what it earns keeping its own work online. The consent rules do not
soften because a machine signed them, and the reference hub asks an
agent public attestation to name a responsible operator key so somebody
is reachable when a notice arrives.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01SKAohrRkqLKVQL2cGCAkR5
* Fix CI: number the PRD requirements and advance the next-id assertion
Two checks the new PRD tripped, both by existing rather than by being
wrong.
The collection validator wants requirements as numbered R# entries and
0006 used a plain ordered list, so it reported OP-L-NO-REQUIREMENTS.
Rewritten as R1 to R7 with priorities, one capability per entry, and the
implementation tracking moved to a paragraph under them where it is not
pretending to be a requirement.
The MCP standards test asserts what the next free PRD id is, and its own
comment says that advances with every PRD added. Adding 0006 makes it
0007.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01SKAohrRkqLKVQL2cGCAkR5
---------
Co-authored-by: Claude Fable 5.1 <noreply@anthropic.com>
BitTorrent distributes bytes better than any CDN and has never been a
product: nobody is paid to seed, and nothing in it is private. Every
Profullstack media property answers that with a central HTTP proxy and a
pass system bolted on, and pays for every byte it serves.
OpenSwarm is an add-on to BitTorrent, carried as BEP 10 extension
messages, that fixes both. The swarm carries AES-256-CTR ciphertext whose
integrity is pinned by two SHA-256 merkle roots in a signed manifest, so
a tracker or DHT node learns an infohash and a size. A leecher buys a
pass over x402 in USDC (the same exchange x402-gateway runs for crawl
passes, settled by CoinPay), a seeder serves inside a bounded credit
window, and the leecher signs a cumulative voucher for every verified
batch. Whoever seeds gets paid. Vanilla clients remain valid members and
browsers remain first-class peers.
Adding a file mints a key pair for it, derived from one publisher seed by
default so there is one thing to back up: the public half is the file's
identity and its BEP 46 key, the private half signs the manifest and
authorises grants and payout changes, and a separate content key
encrypts the bytes and is sealed to paying peers.
The family: a core (records, keys, hashing, transports, discovery,
events), ipfile, ippay, ipdb (a signed hash-chained catalogue with heads
on the DHT), ipaudio, ipvideo, iplive (paid relays with backpressure)
and ipname (Moshpit pins and DNS TXT). Plus the c0mpute.com integration
with seven workload types and thirteen use cases, a proposed ip CLI,
conformance profiles, a security model and an FAQ.
Registered on the site the way OpenCreds is: nav entry, docs registry,
sitemap, a /openswarm landing page, PRD 0005, and the MCP prd_next_id
expectation moved to 0006.
Specs only. No code, no schemas, no reference implementation.
Claude-Session: https://claude.ai/code/session_01YafYxayh7Gqe5MWNNQMev2
Co-authored-by: Claude Fable 5.1 <noreply@anthropic.com>
* Add the LogicSRC OpenCreds specification
Leaving a password manager means writing every secret you own to disk in
the clear, and losing whatever the spreadsheet had no column for. A CSV
is plaintext by construction, lossy by omission, and carries no
integrity: nothing in it says which rows were meant to be there, so a
truncated import looks exactly like a complete one.
The same gap showed up inside LogicSRC. `logicsrc credentials` moves
.env secrets and SSH keys through end-to-end-encrypted team vaults, but
it can only model a key/value pair. A card, a passport, a login with a
TOTP seed, or an OAuth account with a refresh token are all things
people already keep in a vault, and none of them are a key/value pair.
OpenCreds defines three things: the item, the vault, and the database.
- Six item types (login, card, identity, note, key, account) as one
record with a type and a named field group, so everything the user
typed lives in a single encrypted blob. Codes 1-4 match MarkSyncr's
deployed vault and are not renumbered; compatibility is cheaper than
elegance.
- AES-256-GCM over that record with the item id bound as AAD. Without
it, anyone with storage write access could move a low-value login's
ciphertext into a high-value row and watch what the user does next.
- A key hierarchy where the user key is random, not derived, so a
password change re-wraps 32 bytes rather than re-encrypting a vault.
The auth hash comes out of a different HKDF label than the wrap key,
which is what lets it reach a server at all.
- A portable .opencreds file, encrypted by default, whose header is the
AAD over the payload -- so the manifest is authenticated by the same
tag as the data and a truncated import fails rather than reporting
success. The plaintext form exists because people move to products
that read nothing else; it is opt-in, confirmed, 0600, and labelled
"protected": false in its own header.
Namespaces are carried as data, not fixed by the spec: labels are
compiled into every ciphertext a vault has written, so editing one does
not migrate a vault, it makes it undecryptable. MarkSyncr's deployed
vault is conformant by declaring `marksyncr`.
Ships: prd/0004, nine spec pages under docs/opencreds/, six JSON
Schemas, the @logicsrc/opencreds reference implementation with CSV
importers for five products, `logicsrc vault` and the standalone
`opencreds` binary, and the spec page at logicsrc.com/opencreds.
`vault` rather than `creds` because `creds` is already an alias of
`logicsrc credentials`, and the two are different: one moves a pair
between providers, the other stores a record.
@logicsrc/validators now registers every schema by $id before
compiling, so the database schema can $ref the item and manifest
schemas rather than restating them.
120 tests, including CLI end-to-end coverage of the masking rules,
exit codes, and the manifest-mismatch path.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01QRQrfuwuYKKV5UB9kLHuX5
* Make the OpenCreds conformance claim executable
The conformance page described a fixture suite and an `opencreds
conformance` command that did not exist. A specification that documents
a conformance surface it cannot run is a specification nobody can hold
to, including us.
`opencreds conformance` now runs the requirement list as code -- one
check per C-number, carrying its own id and level -- and emits the
report shape the spec publishes. It exits 2 when a MUST does not pass,
so it can gate CI directly. The reference implementation reports 29
passed, 0 failed, 1 skipped; the skip is C19, because key management for
the team profile lives in @logicsrc/plugin-credential-sharing rather
than in this package, and a skipped MAY does not affect conformance.
Fixtures are generated (`--emit-fixtures <dir>`) rather than
hand-written. A vector produced by an implementation and then verified
by it is worth more than a JSON file someone typed: the typed file
drifts silently when the format moves, and the generated one cannot.
Fourteen files, including an invalid/ set every conforming reader must
reject -- a wrong field group, a weak KDF, an unregistered namespace, a
short payload and a tampered manifest.
The CLI requirements stay with the end-to-end tests that drive the real
binary through a child process; a command cannot meaningfully check its
own exit codes, and a masked value that is only masked in the library is
not masked.
conformance.md and cli.md now describe what ships.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01QRQrfuwuYKKV5UB9kLHuX5
* Add @logicsrc/opencreds to the lockfile
`npm ci` refuses a lockfile that does not match package.json, and the
new workspace package plus the CLI's dependency on it were never
recorded: the worktree was bootstrapped by hardlinking node_modules
rather than installing, so npm was never asked to update the lock.
Adds the workspace link and the package entry. No dependency versions
move.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01QRQrfuwuYKKV5UB9kLHuX5
* Register PRD 0004, and stop the fixtures looking like real secrets
Two CI failures, both mine.
`prd/README.md` is generated by `logicsrc prd index --write` and the
scaffold test asserts it is current, so adding a PRD without
regenerating it leaves the repo's own conformance check failing.
Regenerated. The MCP test asserts the next free PRD id against the live
prd/ directory — its comment says it advances with every PRD added — so
it moves to 0005.
ThreatCrush flagged three of the example strings: a PEM header in the
item-model docs and in the conformance fixture, and an `sk_live_`
prefixed token. All placeholders, none real, but the finding is the
scanner working. A fixture only has to exercise the field, and a
real-looking private key header or live-key prefix sitting in the tree
trains both the scanner and the people reading its output to shrug at
exactly the shape that matters. Replaced with obvious placeholders
rather than suppressing the rule.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01QRQrfuwuYKKV5UB9kLHuX5
---------
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
* Add SSH keys and config to credential sharing
Private keys have lived as plaintext-on-disk files guarded only by a
passphrase. This puts them in the same end-to-end-encrypted vaults as
.env secrets, and adds an agent path so a machine can use a key without
ever writing one to its disk.
- `ssh` provider: ~/.ssh as a value bag. Files are picked by sniffing
contents (PRIVATE KEY blocks, ssh-*/ecdsa-*/sk-* public keys) plus
config, config.d/* and allowed_signers. known_hosts and
authorized_keys are host-specific and access-granting, so they need
an explicit --include.
- Each file is one secret carrying a JSON envelope of path, mode and
body. The engine only hands write() the secrets that CHANGED, so a
separate manifest secret would be absent whenever a key's contents
change but the file list doesn't — self-describing values keep every
restore total.
- `logicsrc secrets ssh push|pull|list|agent`, addressed by PERSON not
project: the vault is ssh--<username>, which teams vaults reads as
project ssh, env <username>. One teammate's keys never land in
another's restore; sharing stays a deliberate teams grant.
- Both directions hold back anything that would overwrite a file that
already differs, and say what they skipped. --force opts in. A
restore onto a machine with its own keys is otherwise a way to lose
them.
- Restores chmod each file back to its recorded mode; writeFileSync's
mode applies only on create, so an existing world-readable key would
otherwise stay world-readable. The adapter declares delete:false.
- push warns about passphrase-less private keys before they go up.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
* Add worked examples to secrets and secrets ssh help
Commander's usage line shows only the first alias, so `logicsrc secrets`
— the spelling people actually type — was invisible in its own help.
The examples carry it, alongside the flows worth copying: link/up/down,
the ssh backup round trip, and a plan → dry-run → approve sync.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
* Advertise the ssh provider on the marketing page
The marketing-drift contract failed the build because `ssh` shipped in the
provider registry with no entry in MARKETING_PROOF -- which is the test
working: it exists so a provider cannot ship while the pages people
actually land on still describe the tool without it.
The proof regex is `/~\/\.ssh|SSH key/` rather than a bare `/SSH/` on
purpose. The provider grid renders every registry `name`, and this one is
"Local SSH directory", so `/SSH/` would already be satisfied by the
generated grid and the provider could ship with no copy written about it
at all -- passing the test while failing its intent. Requiring the path or
the phrase means a human wrote a sentence.
That sentence is the new block in the credential-sharing band: ~/.ssh is a
directory of files whose permission bits are load-bearing, not a set of
KEY=VALUE lines, which is the part that makes this provider different from
the other six. README already named ~/.ssh keys, so it needed no change.
apps/logicsrc-web: 75/75 contract tests pass (was 74 passed, 1 failed).
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
* Add the LogicSRC OpenContext specification
OpenContext is an open specification for durable, portable, permissioned,
provenance-aware context shared between humans and AI agents. It defines how
organizational knowledge is described, authorized, versioned, resolved,
audited, and handed between replaceable workers without losing institutional
state.
Follows the OpenPRD/OpenOntology pattern already in the repo: self-contained
JSON Schemas in @logicsrc/schemas, a reference implementation package, CLI
subcommands, docs, examples, and an OpenPRD record.
Schemas (8, all self-contained so a third party can fetch one file and
validate against it with no further resolution):
manifest, object, bundle, role, provenance, decision, diagnostic,
audit-event — registered in @logicsrc/validators and schemas:validate.
Reference implementation (@logicsrc/opencontext):
loader with upward manifest discovery, the full resolution pipeline,
authority/supersession, permissions, redaction, lifecycle, provenance,
deterministic digests, doctor, search, graph, history/diff, guarded writes,
audit events, and file/http/git/sqlite adapters.
CLI: all 15 specified commands, as a standalone `opencontext` binary and as
`logicsrc context`, sharing one implementation so the two cannot drift.
Design decisions worth noting:
- Supersession is declared, never inferred from version numbers. Inferring it
would hide the governance failure it represents and make
multiple-active-versions and duplicate-canonical impossible to detect.
- The bundle digest identifies the resolved context, not the moment it was
computed, so generated_at/bundle_id/digest/as_of are excluded while objects,
lifecycle states, exclusions and warnings are covered. That is what lets a
decision record cite exactly the context that produced it.
- A role's own max_classification beats an inherited one, so a ceiling on a
shared base role cannot silently cap a role deliberately granted more;
requesting several roles at once still takes the lowest, so combining roles
never escalates.
- Scope wildcards match whole dotted segments only. A trailing .* covers a
subtree; an interior * matches exactly one segment. Substring matching here
would be an access-control bug.
- --include narrows an existing scope and is applied after it, never merged
into it, so a request can never widen what a role holds.
Verified: 226 tests across core primitives, permissions/redaction, the
resolution pipeline, security, the published conformance fixtures (13 valid,
35 invalid, 8 resolution scenarios), project behaviour, and the five shipped
examples — which are held to --strict and a 100% health score. Benchmarks meet
every published budget (resolve 1,000 objects in ~33ms against a 2s target).
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
* Point install docs at @logicsrc/opencontext; record the npm name collision
The unscoped `opencontext` name is already published on npm by an unrelated
third party (federicodeponte/opencontext, 2.0.0), so `npx opencontext` would
install a stranger's package. Docs now use `npx @logicsrc/opencontext`; the bin
stays named `opencontext` so the command reads as the PRD specifies once
installed.
Recorded in PRD 0003 as a blocker to resolve before any publication, along with
the fact that no @logicsrc spec package has ever been published, so there is no
existing release path to slot into.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
* Advance the logicsrc-mcp next-PRD-id assertion to 0004
standards.test.ts asserts prd_next_id against the live prd/ directory, so
adding PRD 0003 makes the next free id 0004. The test's own comment
anticipates this: "advances with every PRD added".
Caught by CI, not locally — the earlier verification ran per-package tests for
the packages this branch touches, and logicsrc-mcp is coupled to the PRD
directory without importing from it.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
The credential store resolved its base directory against process.cwd().
Running the CLI from inside a git checkout wrote `.logicsrc/credentials`
into that repo's working tree — a directory containing `vault/`, the one
place raw credential values touch disk — untracked, unignored, and one
`git add -A` from being committed. Two such directories were sitting in
unrelated repos on the machine this was found on.
A per-directory store is also the wrong shape for what the store is for.
It is the record of what was rotated and what the prior values were, and
a record that forks per project folder is several records that disagree.
There is one user, one identity, one vault.
Everything now hangs off a single logicsrcHome(): $LOGICSRC_HOME, else
$XDG_CONFIG_HOME/logicsrc, else ~/.config/logicsrc. The credential store,
the identity and the CLI config all read it rather than each deriving
their own answer — three separate derivations is how the vault ended up
somewhere the config never was.
~/.logicsrc is migrated rather than abandoned. It holds the X25519 secret
key, and losing that loses access to every team vault the member was ever
given, so it is moved on first use; a move that fails says so on stderr
instead of leaving someone silently logged out with a key still on disk
somewhere they were not told about. If the new directory already exists
it wins and the old one is left untouched, because two directories both
claiming to be the identity is how a login writes one and a read finds
the other.
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
Two additions to Credential Sharing.
`logicsrc credentials rotate` (alias `logicsrc secrets rotate`) re-keys a
team vault: fresh DEK, re-sealed to the members who keep access, every
secret re-encrypted under it. Values do not change, so nothing that
consumes them breaks; what changes is that every wrapped key issued
before the rotation is dead. --active (the default) keeps only active
members and revokes the rest -- the "someone left" rotation. --all keeps
everyone who holds access, for plain hygiene. Dry run by default, like
`sync`.
The DEK is recoverable ONLY through the grants, so a half-applied
rotation makes a vault permanently unreadable by everyone. The whole next
state therefore goes to the server in one request and commits in one
transaction (new db.batch helper). The server also requires every
submitted fingerprint to equal the stored one: it cannot see values, but
it can prove a re-key did not swap any. Rotations that would leave the
caller ungranted, grant nobody, or cover the wrong secret count are
rejected before anything is written. GET /vaults/:id/grants now returns
publicKey and status so a client can re-seal in one pass instead of N+1
user lookups, and revocation finally deletes the grant row rather than
leaving one that reports access it no longer confers.
The sh1pt adapter is the fifth provider. It is the only one driven
through a CLI rather than HTTP, because sh1pt publishes
`sh1pt secret set|get|list|rm` as the interface to its vault and
documents no REST endpoint. Values go over the child's stdin, never argv
-- a secret in argv is readable by any user on the host via ps. Since
`sh1pt secret get` needs interactive confirmation it cannot be scripted,
so the adapter is write-only for values like github-secrets: a sync
target, never a source, no value-restoring rollback.
Tests drive a real fake sh1pt binary rather than a mocked execFile, which
is how the hang surfaced: with nothing to pipe, stdin was left open and
any subcommand that reads it would wait forever. It is now always closed.
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
#109 addressed vaults as <project>/<env> and shipped broken: every push
failed with
Vault name must be lowercase letters, numbers, and dashes.
The vault-create endpoint slugifies through /^[a-z0-9][a-z0-9-]{0,62}$/
(apps/pwa/src/routes/credshare.mjs), so a "/" join is refused outright.
Nothing in the CLI ever saw it, because the tests exercised vaultName and
splitVaultName in isolation and never made a request — the one assumption
that mattered, that the server takes an arbitrary vault name, was the one
left unverified.
Switches the separator to "--", which is inside the allowed character set
and still splits unambiguously since neither half may contain one. A
single dash would not: "a-b" + "c" and "a" + "b-c" would collide.
Also validates the joined name against the server's own regex before the
request, so a bad name fails locally with a useful message rather than a
422 after the .env has been read.
The tests now assert the produced name matches that regex, so the
separator cannot drift back out of the allowed set without failing.
Verified end to end against app.logicsrc.com: push, then pull into a
scratch file and diff — keys and values both round-trip losslessly. Then
49 repos pushed under the profullstack team; server reports 49 vaults,
169 secrets, 0 failures.
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
`teams push|pull|grant` took a single `<vault>` name, so a team holding
more than one project had to encode both halves by hand and hope
everyone spelled it the same way. They now take `<project> <env>` and
join them into the `project/env` vault name.
The split lives entirely in the CLI — vaultName()/splitVaultName() are
the only things that know about it, and the server still stores one
opaque vault name — so there's no migration. Both halves reject a "/"
so the join stays unambiguous and the split is a true inverse.
`teams vaults` now breaks the name back into project/env columns,
falling back to the raw name for vaults created before the convention.
Those legacy vaults are no longer addressable (their names don't
contain a slash), so resolveVaultId() lists what the team actually has
instead of just saying "not found" — better than silently retargeting a
push, which in a secrets tool would write to the wrong vault.
Note push/pull carry two different "env"s: the `<env>` positional is
the environment half of the address, `--env` is the local .env path.
Verified commander keeps them separate.
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
