agentleFS
George-RDGeorge-RD/openspine/AGENTS.md

openspine

AGENTS.md2 starsChanged 21 days agoApache-2.0View on GitHub

Rust workspace (crates/) for a self-hostable governed-agent runtime, plus historical OpenSpec specs and archived changes (openspec/) and an Astro docs site (site/). Process canon — read first: DIRECTION.md. New work uses the Pocock skills workflow (wayfinder → grilling/domain-modeling → to-spec → to-tickets → implement, all under .agents/skills/). The retired OpenSpec ceremony is unavailable: never use it for delivery or create a new openspec/changes/ entry. Domain vocabulary: CONTEXT.md (five promises, six users, decision test) and STORIES.md. Crates are split by trust…

The file

Files
AGENTS.md

openspine

Rust workspace (crates/) for a self-hostable governed-agent runtime, plus historical OpenSpec specs and archived changes (openspec/) and an Astro docs site (site/).

Process canon — read first: DIRECTION.md. New work uses the Pocock skills workflow (wayfinder → grilling/domain-modeling → to-spec → to-tickets → implement, all under .agents/skills/). The retired OpenSpec ceremony is unavailable: never use it for delivery or create a new openspec/changes/ entry. Domain vocabulary: CONTEXT.md (five promises, six users, decision test) and STORIES.md.

Crates are split by trust boundary, not by convenience: openspine-schemas (shared types) → openspine-authority (composition) → openspine-gate (mediation) → openspine-kernel (privileged) → openspine-shell (untrusted, sandboxed). Dependencies point one way down that list.

Gate

scripts/check.sh is the gate and mirrors .github/workflows/ci.yml. It strict-validates the canonical historical OpenSpec specs; archived change packages remain static history. Inner loop: cargo test -p <crate>.

Gotchas

  • The kernel E2E test spawns the real openspine-shell binary. cargo test builds shell tests but not the shell binary, so a bare cargo test --workspace fails with a confusing spawn error. Run cargo build -p openspine-shell --bin openspine-shell first (check.sh does this for you).
  • 500-line cap on crates/**/*.rs (scripts/check-file-sizes.sh). The escape hatch is a first non-blank line reading // openspine:allow-large-module reason: <reason>. Prefer the repo's other habit: move the test module into <name>_tests.rs and pull it in with #[cfg(test)] #[path = "<name>_tests.rs"] mod tests;.
  • capabilities/capability-map.json is generated into the public roadmap. scripts/capability-map.mjs rewrites the block between <!-- capability-map:start --> / <!-- capability-map:end --> in site/src/content/docs/roadmap.md. Edit the JSON, never inside the markers.
  • Every test: <name> row in docs/threat-claims.md must name a real test (scripts/check-claims.sh). Writing a security claim without landing its test fails the gate.
  • OpenSpec is historical spec validation only. It is pinned to exact 1.6.0-beta.1; scripts/check.sh uses the local binary solely to run validate --all --strict over openspec/specs/. Archived change packages are not revalidated. Do not run the change ceremony. Rust is pinned to 1.97.1 in rust-toolchain.toml.
  • Requirement history lives in the canon. .raw/openspine-agentos-design-log.md and .raw/openspine-decision-log.md record settled product decisions; the archived OpenSpec tree records their former specification shape. Current process comes from DIRECTION.md and the issue tracker, never from the historical OpenSpec sequence.

Where to look

  • Process rulings: DIRECTION.md. PR and stack mechanics: docs/agents/delivery.md.
  • Operator-facing docs: docs/ (terminal chat, Gmail setup, day-2 ops, threat claims).
  • Product framing and the public site: site/.
  • Whole-codebase structural view: the codegraph MCP server, registered in .omp/mcp.json (omo-codegraph), so omp sessions here get codegraph_explore / codegraph_search / codegraph_callers / codegraph_node as tools. explore gives a symbol's blast radius, callers, and covering tests in one shot — use it instead of reading source file-by-file. The index is a local .codegraph/codegraph.db (~34MB, gitignored). It is NOT built on demand: run ~/.omo/codegraph/bin/codegraph index once (~5s; the binary is not on PATH), after which the MCP file-watcher auto-syncs on edits. Re-run index after a branch switch or large rebase if results look stale.

Agent skills

Portable engineering skills and their support files are versioned under .agents/skills/; skills-lock.json records their sources. OMP-only extras under .omp/skills/ are supplemental and do not define delivery.

Issue tracker

Issues, specs and the roadmap live in Linear: the OpenSpine project (P-ANO-23) of team ANO, managed with the Linear MCP tools. GitHub Issues are closed history; historical #N references stay valid. See docs/agents/issue-tracker.md.

Triage labels

Four of the five canonical triage roles are labels, used verbatim; wontfix maps to the Canceled state. See docs/agents/triage-labels.md.

Domain docs

Single-context: CONTEXT.md + docs/adr/ at the repo root. See docs/agents/domain.md.

Delivery

Code lands on main through pull requests. One issue may use a short stack of complete, tested, genuinely dependent layers. Review every layer against standards and spec, then independently review the cumulative stack head. See docs/agents/delivery.md.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.