agentleFS
mhxiemhxie/Atelier/.agents/skills/atelier/SKILL.md

atelier

Skill5 starsChanged 36 days agoMITView on GitHub

Run or modify Atelier skills, routines, agents, tools, intent routing, and harness portability in this repo. Use for broad Atelier harness work or to adapt Claude `/hi` to native Codex `$hi`.

12 other projects use a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

6 rules. 0 are common, 1 is distinctive, and 5 are very specific to this project.

  • 0

    Common

    Most of the 12 edited versions we read keep these.

  • 1

    Distinctive

    This author's own choices. Worth a look.

  • 5

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Use this skill when the user asks to run or modify Atelier workflows, components, or harness portability.

    line 8
  2. Distinctive:

    Read AGENTS.md for the shared contract, runtime adaptations, and harness checklist.

    line 14
  3. Very specific:

    Read protocols/runtime-adapters.md only when changing or debugging cross-runtime behavior.

    line 16
  4. Very specific:

    Do not launch Codex recursively, and do not launch bot-invoked workflows such as autoevo-nightly from shorthand.

    line 24
  5. Very specific:

    Follow the checklist in AGENTS.md and harness/README.md.

    line 97
  6. Very specific:

    Before edits, read protocols/repo-conventions.md → Editing discipline for whole-feature budgets, approval thresholds, smoke scope, and cost …

    line 108

“Common” is measured: kept by at least 6 in 10 of the 12 edited versions of this file we could read. “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
SKILL.md

Atelier

Use this skill when the user asks to run or modify Atelier workflows, components, or harness portability. It is the one hand-written Codex edge file; it points at the canonical sources and adds only what is Codex-native.

Quick Start

  1. Read AGENTS.md for the shared contract, runtime adaptations, and harness checklist. It is the single instruction source for Claude Code and Codex.
  2. Read protocols/runtime-adapters.md only when changing or debugging cross-runtime behavior.
  3. Invoke known workflows through their explicit repo skills ($hi, $weekly, $review, $triage, $lint, and so on). Each reads the matching skills/<name>/SKILL.md source and runs it in the current thread. For $hi, classify the request against scripts/intent_coverage.py catalog and read only the selected row's procedure. No fit is a semantic handoff through intents.general, never implicit reflection.
  4. Do not launch Codex recursively, and do not launch bot-invoked workflows such as autoevo-nightly from shorthand.
  5. Discover native roles under .codex/agents/ and inspect them with /agent. Agent(...) in a command means dispatch the matching project agent; if dispatch is unavailable, run the role sequentially from agents/<role>.md and disclose the downgrade.
  6. For external launches, scripts/atelier_runtime.py resolves the committed Codex default from harness/runtimes.toml, the gitignored local preference, and one-process overrides.

Codex-native notes

  • Codex CLI slash input is reserved for built-in TUI commands; project commands are $name skills with allow_implicit_invocation: false. $reflect runs $hi. Translate /name references to $name; never translate real Codex built-ins such as /hooks or /agent.
  • Read is the local file, Grep/Glob are rg and rg --files, Bash is the local shell, AskUserQuestion is a native choice UI or a concise numbered question.
  • Type $ in the Codex composer for command discovery. Do not guess an unavailable command.

Operations

  • Local retrieval uses scripts/semantic.py: Reflect's lexical index first, then the pinned QMD dependency. sources/semantic.md owns setup, scopes, and the bounded JSON contract. Queries use cached models only; downloads require explicit initialization.
  • Vault title/rendering hygiene: scripts/zk_audit.py; naming: protocols/repo-conventions.md. Linked ledger tasks use scripts/todos.py check and sync --file <quarter>.md (preview; --apply writes); see protocols/local-first-architecture.md.
  • Wiki articles: protocols/wiki-schema.md governs creation, edits and metadata-driven localized titles; scripts/trust.py reads canonical identities, bounded claims and evidence. Curator preserves identity and review history; Forgetter reads bounded claims. L1–L4 labels sync through zk_audit.py --sync-reflect-levels; protocols/local-first-architecture.md owns the contract.
  • Route context packs selected repository profiles and registered session logs with pinned Repomix; protocols/session-continuity.md owns selection and ceilings.
  • Reflection's energy and exploration intents share skills/daily-reflection/SKILL.md while keeping their selected context. That procedure and protocols/session-log.md own the branches and compact logs.
  • $digest uses collect --json then write into one Reflect-native note; the scheduled daily-digest routine uses configured CodexBar sources and compact quota tables. The shared command owns setup, offline boundaries, approval before a rewrite, and the three-edition overdue TODO limit; never sync during a run.
  • Reading feedback connects Curate, Read, Introspect, and explicit policy evaluation through protocols/decision-ledger.md → Reading feedback loop. Use its typed decisions.py helpers for compact event batches and stored policy IDs; load Offline reading evaluation only for policy experiments.
  • Dispatch inputs and completion reporting live in protocols/agent-handoff.md; load its common sections and the selected payload contract. The procedure owns calls; protocols/agent-handoff.md also owns responsibilities and voice legs. Reader owns the shared reading behavior; Scholar keeps only its role settings and the shared-contract pointer.
  • Native observation and routine reports: sources/runtimes/observability.md. Direct API calls are unlogged.
  • For runtime capability maintenance, read sources/runtimes/README.md and the registry's references; CLI discovery does not prove activation.
  • Local scheduled routines: validate declarations with uv run --frozen python scripts/routine_prefect.py validate --json and read recent Prefect state with uv run --frozen python scripts/routine_status.py. scripts/cron_spec.py uses Prefect's cron engine for health-check dates and cadence. The transfer procedure is in scripts/launchd/README.md: stop every source scheduler before loading the Prefect deployment service. Sandbox path handling lives in scripts/routine_adapter.py; runtime ownership follows protocols/runtime-adapters.md. Artifact evidence and shared receipt validation live in protocols/remote-routines.md. Connector discovery before reporting missing inputs follows routines/_adapters/archived-prompt/PROCEDURE.md. Autoevo drafts only in its isolated workspace; the trusted parent owns its plain-file vault writes and structured result; notes require applied changes or fresh triage (protocols/autoevo.md).
  • Private routine mappings, tools, routines, digest declarations, and skills stay under their registered $OV roots. Classification and activation live in protocols/components.md; remote execution and digest context live in protocols/remote-routines.md.
  • $civ loads framework definitions from the private source referenced by the profile.

Harness Changes

Follow the checklist in AGENTS.md and harness/README.md. The registries are harness/skills.toml, harness/agents.toml, harness/intents.toml, harness/models.toml, harness/capabilities.toml, harness/retrieval.toml, harness/paths.toml, and harness/runtimes.toml, and routines/registry.toml; edit them, never the generated runtime edges. Run python3 scripts/harness_lint.py before finishing and scripts/harness_smoke.py after helper or registry edits. Keep runtime edges thin: they point to canonical sources and do not copy workflow bodies. Public configuration shape lives in harness/registry.schema.json; lint uses the pinned validator in the project .venv. Setup and error codes are documented in harness/README.md; lint never installs missing dependencies itself.

Before edits, read protocols/repo-conventions.md → Editing discipline for whole-feature budgets, approval thresholds, smoke scope, and cost reporting.

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.