agentleFS
emersonnakashimaemersonnakashima/task-spec/AGENTS.md

task-spec

AGENTS.md0 starsChanged 2 months agoMITView on GitHub

Before any task, read OPERATING.md. Do not invent a second loop. Do not skip the HMAC seal. Do not put family hops in docs/. This repo is the task-spec engine: an open, vendor-neutral, self-verifying unit-of-work format (markdown + YAML frontmatter + runnable bash evals) plus the bash toolchain that authors, gates, dispatches, and accepts those tasks. It was extracted from converge/skills/task-spec at v3.3.0 and now stands alone. Frozen artifacts under release/<version>/, .taskspec/acceptance/, and tasks/done/ describe what actually shipped. Never rewrite…

  • Where files are headingNew AGENTS.md files have more rules about keeping secrets out of the code, and fewer about code style.

    Of the AGENTS.md files we read, the 32,671 first seen in the last six months, against the older ones: the kind of rule that gained most share, and the kind that lost most. About files like this in general, not this one, and an estimate, since a kind is a pattern in the words.

  • In this repository3 other files it gives its agents.

The file

We split this AGENTS.md into the rules it gives an agent

7 rules. About 0 are common, 6 are distinctive, and 1 is very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈6

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 1

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Before any task, read OPERATING.md.

    line 3
  2. Distinctive:

    Do not invent a second loop.

    line 4
  3. Distinctive:

    Do not skip the HMAC seal.

    line 5
  4. Distinctive:

    Do not put family hops in docs/.

    line 6
  5. Distinctive:

    All of the above must be green before a change is considered done.

    line 68
  6. Very specific:

    bash-3.2 floor on the core gate path: No bash-4-only constructs (declare -A, mapfile/readarray) in src/gate/, src/accept/, or src/lib/.

    line 72
  7. Distinctive:

    Keep edits minimal and in the style of the surrounding file.

    line 86

≈ “Common” is an estimate: at least 1 in 5 of the 41,581 AGENTS.md files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
AGENTS.md

Agent entry

Before any task, read OPERATING.md. Do not invent a second loop. Do not skip the HMAC seal. Do not put family hops in docs/.

AGENTS.md — contributing to the task-spec engine

This repo is the task-spec engine: an open, vendor-neutral, self-verifying unit-of-work format (markdown + YAML frontmatter + runnable bash evals) plus the bash toolchain that authors, gates, dispatches, and accepts those tasks. It was extracted from converge/skills/task-spec at v3.3.0 and now stands alone.

Repository layout

PathHolds
bin/taskspecthe one CLI entry point; dispatches to src/
spec/normative: the v3/v4 format, JSON schemas, conformance suite
interop/UPSTREAM.lock — protocol version/commit/digest lock. Path is frozen by 3.8.1 evidence; do not move it
src/Python + Bash engine, including native decompose/, signed recipe/, CLI metadata, and src/templates/
src/meshctl/Python TaskMesh cockpit (taskspec mesh); not the Go daemon
mesh/the optional Go control plane (mesh/cmd/, mesh/internal/mesh/)
harness/non-normative host surfaces; see harness/README.md for the inclusion rule
tests/the self-test suite, all fixtures under tests/fixtures/, benchmark cases in tests/evals/
release/shipped runtime inputs and frozen per-version evidence; see release/README.md
docs/the knowledge base; start at docs/index.md. New how-tos go in docs/guides/; docs/runbooks/ is closed
tools/repository scripts: release builders, renderers, signing-key provisioner
tasks/, .taskspec/this repo's own dogfooded backlog and acceptance receipts — not the public tutorial; see tasks/README.md
skills/the in-repo skill pack; skills/task-spec/SKILL.md is a byte-for-byte copy of root SKILL.md, enforced by tests/test-repo-layout.sh
assets/README images only
.claude-plugin/marketplace source of truth for the Claude plugin
.github/workflows and dependabot config
CONTRIBUTING.md, SECURITY.mdhuman entry points; agents still follow this file

Frozen artifacts under release/<version>/, .taskspec/acceptance/, and tasks/done/ describe what actually shipped. Never rewrite them to match a later refactor.

tests/test-repo-layout.sh enforces this table: the set of top-level directories and root files is declared there, and adding either one requires updating the test and this table in the same change. That is deliberate — it is what keeps the tree from drifting back into twenty-odd top-level directories.

Build / test

There is no build step — the engine is bash + markdown. The single release gate (hosted CI runs exactly this on Ubuntu and macOS; run it on macOS locally before touching the bash-3.2 gate path):

make check            # doctor + lints + all self-tests + conformance

Or run the pieces individually:

bash bin/taskspec doctor                       # toolchain sanity (bash, git, crypto, key)
bash tests/lint-skill-docs.sh                  # VERSION == CHANGELOG == _lib.sh
bash tests/lint-docs.sh                        # markdown links resolve, fences balanced
for t in tests/test-*.sh; do bash "$t"; done   # full self-test suite
bash spec/conformance/run_conformance.sh       # conformance suite (default self-adapter)
bash bin/taskspec conformance --self-test      # L0–L2 against the bundled ref executor

All of the above must be green before a change is considered done.

Conventions

  • bash-3.2 floor on the core gate path. No bash-4-only constructs (declare -A, mapfile/readarray) in src/gate/, src/accept/, or src/lib/. macOS system bash 3.2.57 must run them. Aux scripts that need bash 4 (e.g. src/backlog/lint-backlog.sh) must call ts_require_bash4. tests/test-bash-portability.sh enforces this.
  • VERSION is the single version source. Bump VERSION, mirror it in src/lib/_lib.sh:TASKSPEC_VERSION, and add a CHANGELOG entry — the lint asserts all three agree.
  • Format changes are triple-locked. Any change to the spec format requires: the JSON Schemas in spec/schemas/ updated, conformance fixtures in spec/conformance/ updated, and a CHANGELOG entry. format_version bumps are MAJOR.
  • Scripts resolve everything relative to their own location (via src/lib/_lib.sh). Never hardcode absolute paths.
  • Keep edits minimal and in the style of the surrounding file.

Boundaries

  • harness/ (engines, trackers, adapters) is non-normative — the contract is spec/ + the conformance suite. Adapters may lag; the spec may not.
  • tests/fixtures/diamond-6/ are CI fixtures: never --stamp or otherwise mutate their frontmatter sigs in tests.
  • No git mutations (commit/push/reset/rebase) unless the user explicitly asks. Leave the working tree for review.

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.