agentleFS
EnjoyoerEnjoyoer/loadout/skills/orchestration/repo-lessons/SKILL.md

repo-lessons

Skill0 starsChanged 27 days agoMITView on GitHub

Maintain an assigned project's or durable workstream's STATUS.html and LESSONS.md: root and ownership rules, the five-section dashboard, and capped durable lessons.

The file

We split this skill into the rules it gives an agent

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

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈1

    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:

    When resuming a workstream or relying on earlier decisions, read the relevant owned state and lessons.

    line 25
  2. Very specific:

    Use $handoff for handoffs.

    line 60

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill 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
SKILL.md

Project Memory

Each owner maintains exactly two memory files in its assigned root:

  • STATUS.html is the fixed current-state dashboard.
  • LESSONS.md is the concise record of durable, non-obvious facts.

Edit both in place. Never write backup copies of them (.bak, dated copies, or archives); Git or the platform's file history is the recovery path.

Root and ownership

Resolve write ownership from the assigned role and root: a PM owns its project-root pair; a durable Worker owns its assigned workstream-root pair. Git is optional. Existing files do not confer ownership. A Worker must not walk upward into a PM's pair or take it over because its own pair is absent. If the assignment leaves the root unclear, clarify before writing memory; unrelated safe work may continue.

Each owner writes only its own pair autonomously. A PM reads relevant Worker files one-way when needed and maintains its own summary. Workers maintain their pair without waiting for the PM or reading/mirroring parent PM state every turn. Memory maintenance requires no callbacks, per-turn completion messages or acknowledgments, task/host registration, owner task links, or coordination gates. Explicit task instructions, shared source context, and normal task communication remain valid.

Create missing files only in the explicitly assigned durable root, using an approved template or a minimal pair. Do not seed unrelated one-off projects or create additional memory trackers, handoff files, or archives. An explicitly assigned replacement resumes the existing pair; absence or staleness never authorizes taking over another owner's memory.

Read once, then maintain

When resuming a workstream or relying on earlier decisions, read the relevant owned state and lessons. Read only the context needed for the current task; refresh when source facts changed or contradict memory.

STATUS.html

STATUS.html is a fixed visual dashboard, subordinate to source evidence: browser-openable, restrained CSS, at most 25,000 bytes. Under the project name it has exactly these five sections, in this order:

  1. Waiting on you: owner decisions, oldest first, each with recommendation and deadline if known.
  2. In flight: each item with its exact known agent, workspace, branch, PR, session, and schedule IDs, its state, and its next action or trigger.
  3. Blocked: each blocker with its dependency and the party it waits on.
  4. Done last 7 days.
  5. Health line: last evidenced check, known gaps, and as-of time.

Rewrite facts in place after a material milestone, blocker or ownership change, closeout, or maintenance correction. Do not put raw logs, secrets, historical narrative, archives, or model-policy notes here. No design skill is required to maintain this operational file.

LESSONS.md

Keep only facts whose loss would likely cause a future mistake or wasted investigation. Add or edit an entry after a non-obvious project-specific discovery, proven workaround, ownership rule, path, tooling fact, or behavior. Do not copy current status, progress, worker profiles, global policy, secrets, or facts already clear in source documentation.

Cap LESSONS.md at 8,192 UTF-8 bytes. Keep the highest-risk actionable lessons first, especially ownership and safety boundaries and demonstrated recurring pitfalls; merge duplicate advice, point to evidence, and drop lower-risk history instead of exceeding the cap.

Durable entries carry Status, Verified, Evidence, and Review when. Put newest entries first. Remove stale, duplicate, or superseded material when it no longer prevents a likely regression. Verified is the last evidence-backed verification date: advance it only when a current check confirms the fact. Failed or unavailable checks leave the recorded date unchanged and set Status to needs-review. Use not verified only for a fact with no established verification; never invent dates.

### YYYY-MM-DD - short title
- **Status:** active | superseded | needs-review
- **Verified:** YYYY-MM-DD | not verified
- **Evidence:** project-relative path, source URL, or commit ID
- **Review when:** concrete event that can invalidate this fact
- **What:** ...
- **Why it matters:** ...
- **Fix / workaround:** ...

Handoff

Use $handoff for handoffs; a successor resumes by reading the pair as in "Read once, then maintain".

Evidence and contradictions

Store evidence as project-relative paths (identify the owning project when needed), source URLs, or commit IDs; resolve paths in the current environment. Keep the portable contract independent of host paths, shells, usernames, model profiles, and platform task handles. Label genuinely platform-specific facts with their applicability so another platform does not inherit them as universal truths. This contract defines no cross-machine sync or transport; unavailable evidence remains unverified.

Live source/config and successful harmless checks outrank notes; STATUS.html outranks LESSONS.md, which outranks transcript memory. Before relying on a lesson during non-trivial work, check its evidence and review trigger with a bounded read-only check. Do not mutate from a stale or unverified memory claim.

When evidence conflicts with memory, represent uncertainty and supporting evidence only in your own pair, marking affected owned lessons needs-review. Correct, supersede, or remove owned stale claims once evidence is clear. A PM observing stale or conflicting Worker facts updates its own summary, never the Worker's files, and sends no unsolicited memory-correction callback. The same write boundary protects all third-party PM/Worker memory. Block only decisions depending on unresolved facts; unrelated safe work may continue.

Maintenance boundary

After a material milestone, blocker, ownership change, closeout, or durable discovery, update the affected owned file before handing off. Update both only when both need changes. Refresh the changed file's single Last maintained: <ISO-8601 timestamp with timezone> marker; this records maintenance, not verification. With no material change, write nothing. Report failed required writes separately and still deliver the user result.

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.