aesop / skills
CLAUDE.md2 starsChanged 31 days agoMITView on GitHub
Five orchestration skills, each invoked via CLI (e.g., /buildsystem, /power, /dashboard): Backlog format (state/BACKLOG.md): Time estimates: 3–5 min (typo/config), 5–10 min (function/test), 10–15 min (module refactor), 15+ min (split smaller). Work-order contract (each agent receives): - Task: one backlog item (feature/fix/tech debt) - Agent type: backend-dev, frontend-dev, test-bot, docs-agent, etc. - Time budget: 3–10 min (Haiku best-effort scope) - Scope: reproduce, change location, test command, commit message template - Constraints: never modify CLAUDE.md/STATE.md, must pass pre-commit hook, no external calls
In this repository21 other files it gives its agents.
- auth/CLAUDE.mdCLAUDE.md
- bin/CLAUDE.mdCLAUDE.md
- chat/CLAUDE.mdCLAUDE.md
- CLAUDE.mdCLAUDE.md
- daemons/CLAUDE.mdCLAUDE.md
- dash/CLAUDE.mdCLAUDE.md
- driver/CLAUDE.mdCLAUDE.md
- hooks/CLAUDE.mdCLAUDE.md
- mcp/CLAUDE.mdCLAUDE.md
- monitor/CLAUDE.mdCLAUDE.md
- scan/CLAUDE.mdCLAUDE.md
- state_store/CLAUDE.mdCLAUDE.md
- stdio/CLAUDE.mdCLAUDE.md
- tests/CLAUDE.mdCLAUDE.md
- tools/CLAUDE.mdCLAUDE.md
- ui/CLAUDE.mdCLAUDE.md
- buildsystemSkill
- dashboardSkill
- fleetSkill
- healthcheckSkill
- powerSkill
The file
We split this CLAUDE.md into the rules it gives an agent
4 rules. About 0 are common, 4 are distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈4
Distinctive · estimate
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 42
No two backlog items may own the same file or directory
- Distinctive: line 60
All PRs from a wave merge through serial merge train
- Distinctive: line 95
buildsystem: Run a full wave (in a test repo with mock backlog)
- Distinctive: line 106
healthcheck: Check fleet health
≈ “Common” is an estimate: at least 1 in 5 of the 27,641 CLAUDE.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.
Files
skills/ — Orchestration and priming skills
Universal rules (every domain)
- Feature branch only, never main; every push gated by
python tools/secret_scan.py --stagedexit 0. - Tests never pollute cwd or global git config; temp dirs only; dummy secrets are runtime-concatenated, never literal.
- In worktrees use ABSOLUTE paths under the worktree for every write.
- Domain docs stay minimal-but-complete; update this file in the same PR as code it describes.
Module layout
Five orchestration skills, each invoked via CLI (e.g., /buildsystem, /power, /dashboard):
- power/ — prime the filesystem brain (load CLAUDE.md rules, dispatch model, memory, state); init-prime or fast-path for already-primed repos
- buildsystem/ — execute one complete orchestrated development wave (backlog rank → parallel Haiku fleet → merge train → checkpoint → audit)
- dashboard/ — launch the aesop web dashboard (idempotent startup, SSE realtime updates, CSRF protection)
- healthcheck/ — check fleet health (one colored ball: green/yellow/red based on heartbeats, alerts, orchestrator status)
- fleet/ — snapshot fleet state as JSON (agents, heartbeats, tracker lanes, orchestrator status) — never mutates
Buildsystem manifest & work-order contract
Backlog format (state/BACKLOG.md):
## P1: Critical
- [ ] Task title — agent-type — time-estimate
Example: Fix secret-scan hang — backend-dev — 5min
## P2: Features
- [ ] Add feature X — frontend-dev — 8min
## P3: Tech Debt
- [ ] Refactor module Y — backend-dev — 15minTime estimates: 3–5 min (typo/config), 5–10 min (function/test), 10–15 min (module refactor), 15+ min (split smaller).
Work-order contract (each agent receives):
- Task: one backlog item (feature/fix/tech debt)
- Agent type: backend-dev, frontend-dev, test-bot, docs-agent, etc.
- Time budget: 3–10 min (Haiku best-effort scope)
- Scope: reproduce, change location, test command, commit message template
- Constraints: never modify CLAUDE.md/STATE.md, must pass pre-commit hook, no external calls
Preflight guard invariant (wave-flat-dispatch.template.mjs):
- No two backlog items may own the same file or directory
- Each item assigned to exactly one agent
- Preflight detects overlap and aborts with remediation plan before dispatch
- Template includes documentation strings (dispatch-ok markers suppress lint false positives on tool references)
Phases:
- Preflight (1–2 min) — validate setup, check heartbeat, parse backlog, detect file overlap
- Backlog & Dispatch (5–10 min) — rank items, assign agents, create work orders
- Fleet Execution (30–90 min) — Haiku agents work in parallel, in isolated worktrees
- Code Review & Integration (10–30 min) — run CI, merge train (serial), flag failures
- Checkpoint & Audit (2–5 min) — update STATE.md, append BUILDLOG.md, commit + push
Worktree isolation (critical):
- Each agent runs in
aesop-wt-<wave>-<agent-id>/(independent branch, HEAD, working dir) - Zero cross-agent interference
- Automatic cleanup after wave closes
Merge serialization (critical):
- All PRs from a wave merge through serial merge train
- First agent to green CI merges, next waits for main to stabilize, re-tests, then merges
Power skill
Two entry paths:
- Init-prime (unprimed repo, no CLAUDE.md): fan-out Haiku explorers → synthesize project CLAUDE.md + domain units + STATE.md seed → commit + push
- Already-primed (fast path): load global rules (
~/.claude/CLAUDE.md), load memory (MEMORY.md), load active project STATE.md, report compact brief, optionally spin up app and monitor background loops
Prerequisites: Aesop scaffolded into repo via CLI (npx @matt82198/aesop . --name "my-project").
Healthcheck skill
One-line status command: python tools/healthcheck.py (or --json for machine-readable).
Output ball:
- 🟢 Green — watchdog <300s, monitor <3600s, no HIGH alerts
- 🟡 Yellow — stale heartbeat OR unreviewed MED alert
- 🔴 Red — HIGH alert OR watchdog dead (>600s) during active dispatch
Fleet skill
One-shot snapshot: aesop fleet (JSON output, read-only, never mutates state).
Returns:
- heartbeats (watchdog, monitor: age, status, thresholds)
- agents (active: id, project, status, age, tokens, task label)
- tracker (backlog counts by lane)
- orchestrator (activity, phase, timestamp)
Graceful degradation: missing files produce unavailable: "<reason>" entries.
Test commands
- buildsystem: Run a full wave (in a test repo with mock backlog)
cd <test-repo> /buildsystem # Verify: STATE.md updated, BUILDLOG.md appended, PRs merged - power: Prime a new repo or already-primed repo
/power # Verify: compact brief output, CLAUDE.md files readable, STATE.md NEXT STEPS clear - healthcheck: Check fleet health
python tools/healthcheck.py python tools/healthcheck.py --json - fleet: Snapshot fleet state
aesop fleet # Verify: JSON valid, heartbeats present, agents or "unavailable" message
Key invariants & gotchas
- Buildsystem is stateless across phases — each phase stands alone (safe to restart mid-wave)
- Backlog must be parseable (YAML or markdown, consistent format) — preflight will abort on parse errors
- File ownership overlap is fatal — preflight guard prevents two agents from touching same file
- Worktrees auto-cleanup — don't manually delete
aesop-wt-*dirs; buildsystem handles it - Power init-prime writes repo files — controlled commit + push via secret-scan gate
- Healthcheck is read-only — diagnostics only, never repairs
- Fleet snapshot is JSON canonical — orchestrators use this, not ad-hoc state reads
Domain-specific details
Each skill has its own SKILL.md file in its directory with complete implementation details, manifests, and troubleshooting:
skills/buildsystem/SKILL.md— phases, manifest, work-order contract, troubleshootingskills/power/SKILL.md— init-prime and fast-path workflowsskills/healthcheck/SKILL.md— diagnostics and repair proceduresskills/fleet/SKILL.md— snapshot structure and queries
Map of all domains: /CLAUDE.md
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
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.

