agentleFS
peterzatpeterzat/OrgSmith/.claude/skills/forge-author/SKILL.md

forge-author

Skill6 starsChanged 33 days agoApache-2.0View on GitHub

Worker for one OrgSmith work order. Reads a work-order JSON, authors the deliverable (persona enrichment or DocIR documents), and ingests it via the orgsmith CLI. Spawned by /forge with a fresh context per batch; not normally invoked by hand.

The file

We split this skill into the rules it gives an agent

11 rules. About 0 are common, 6 are distinctive, and 5 are 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.

  • 5

    Very specific

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

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

    Read the work-order file you were given.

    line 16
  2. Very specific:

    Follow its instructions field exactly.

    line 18
  3. Very specific:

    Write the deliverable JSON to <workorders-dir>/reply-<work-order-id>.json (replace : with - in the id for the filename).

    line 19
  4. Very specific:

    NOT ingest. Write the reply, stamp the generator, and report the reply path back.

    line 41
  5. Distinctive:

    If you ingest and it rejects, the output lists every problem.

    line 44
  6. Distinctive:

    Write sentences so the substituted value reads naturally ("a fixed fee of {{fact:E-2019-001.fee}} covering...").

    line 59
  7. Distinctive:

    No invented names, employers, addresses, amounts, or calendar dates.

    line 62
  8. Very specific:

    Every surface string in a brief's mentions list must appear verbatim in that doc's text (sigblock signers cover themselves).

    line 63
  9. Distinctive:

    Never touch files outside the deliverable you write.

    line 72
  10. Distinctive:

    Never edit ledgers, manifests, state, or rendered files.

    line 73
  11. Distinctive:

    Never re-run pipeline stages other than the single ingest verb.

    line 74

≈ “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

forge-author — one work order in, one validated deliverable in the org

You are an author inside a document-generation pipeline. Your entire job: read ONE work-order JSON file, produce the deliverable it asks for, ingest it, and report one line back.

PY means the project venv python: .venv/bin/python.

Procedure

  1. Read the work-order file you were given. It is self-contained: org narrative, instructions, briefs, and the deliverable schema name. Follow its instructions field exactly; it is the contract.
  2. Write the deliverable JSON to <workorders-dir>/reply-<work-order-id>.json (replace : with - in the id for the filename).
  3. Stamp the deliverable with what actually wrote it — a top-level generator object alongside work_order_id:

    "generator": {"model": "<your exact model id>", "effort": "<your effort>"}

    The model id is the one you are running as, from your own context. The effort is the one your dispatch prompt names; if it names none, use $CLAUDE_EFFORT (echo $CLAUDE_EFFORT), and omit the whole generator object if that is unset too rather than guessing. Report yourself, not the orchestrator: you are what authored these words, and a forked worker does not always inherit the session's model. This is a record for a human to read, never a check — nothing validates it and nothing fails without it.

  4. Deliver. Which of these you do is set by your dispatch instructions:
    • Ingest yourself (enrichment orders, and any order told to ingest):
      • enrichment order: PY -m orgsmith foundation <slug> --ingest <file>
      • authoring order: PY -m orgsmith author <slug> --ingest <file>
    • Author only (authoring batches in /forge's parallel window): do NOT ingest. Write the reply, stamp the generator, and report the reply path back. The orchestrator ingests every author deliverable itself, serially, so concurrent batches cannot corrupt state.json.
  5. If you ingest and it rejects, the output lists every problem. Fix the deliverable and re-ingest, at most 2 retries. If still rejected, report the rejection output verbatim and stop. If you were told to author only, a rejection is the orchestrator's to handle: report the reply path.

Writing quality (authoring orders)

  • The reader must believe a person at this firm wrote it in that year. Match the org narrative's voice; vary sentence rhythm; no template feel; no modern AI-assistant tone. Nothing may sound like the same person wrote every document.
  • Era-appropriate: nothing in the text may postdate the document's date (tools, idioms, events).
  • Placeholders are sacred: every briefed fact id appears as {{fact:<id>}} exactly; you never know or invent the underlying value. Write sentences so the substituted value reads naturally ("a fixed fee of {{fact:E-2019-001.fee}} covering...").
  • People are only those in the briefs, with their exact names and titles. No invented names, employers, addresses, amounts, or calendar dates.
  • Every surface string in a brief's mentions list must appear verbatim in that doc's text (sigblock signers cover themselves). Nickname surfaces go in prose naturally, alongside the person's full name, the way colleagues actually write.
  • Respect genre structure from guidance (letters end with a sigblock; minutes name attendees in a list; tables where briefed).

Hard limits

  • Never touch files outside the deliverable you write.
  • Never edit ledgers, manifests, state, or rendered files.
  • Never re-run pipeline stages other than the single ingest verb.

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.