agentleFS
WingedGuardianWingedGuardian/GENesis-AGI/.claude/skills/deliverable-builder/SKILL.md

deliverable-builder

Skill103 starsChanged 7 days agoMITView on GitHub

Produce a professional, send-ready deliverable on the user's behalf — in the correct file format (never raw markdown), in the user's voice, structured to lead with the strongest material, and free of document-level AI tells. Runs a gated pipeline with a fresh-context verification policeman and a Stop-hook that won't let the session finish without a verified PASS. Use when producing a job take-home, client report, executive one-pager, slide deck, proposal, or anything going out under the user's name.

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:

    Read each reference file when you reach its stage — don't load them all at once.

    line 20
  2. Very specific:

    Stop-hook does NOT apply. See references/intake.md → "Autonomous mode" and references/approval-gates.md.

    line 68

≈ “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

Deliverable Builder

This skill is the conductor. It produces a finished, human-quality deliverable by running a fixed pipeline of stages — and it makes that process non-skippable, because a skipped or lazy stage shows up at Gate 2 as a requirement the artifact fails. Most stages reuse an existing skill or tool; this skill owns the sequence, the gates, and the quality bars.

Read each reference file when you reach its stage — don't load them all at once.

When to use

Any time the user asks you to produce something they'll send — a take-home, client report, one-pager, deck, proposal. Not for internal scratch notes, code (use code-voice), or content you're publishing as Genesis (use content-publish / genesis-voice).

The pipeline

#StageRead / invokeGate
1Intake — frame, freeze acceptance criteria, pick format, open the specreferences/intake.mdGate 1 (interactive)
2Draft — produce the actual substance against the frozen criteria(your analysis)
3Structure & altitude — lead with the answer, action titles, methodology→appendixreferences/structure-altitude.md
4Voice — make it sound like the userinvoke the voice-master skill
5Anti-slop — strip document/sentence AI tellsinvoke the humanizer skill (voice-calibration off — voice-master already voiced it)
6Render — produce the real file (PDF/DOCX/PPTX/XLSX); set status: rendered_unverifiedreferences/render-guide.md (pandoc / /make-pdf / /drawio-skill / OfficeCLI when office_deliverables active)
7Verify — fresh-context adversarial PASS/FAIL vs. the frozen criteriareferences/qa-protocol.mdGate 2 (policeman)
8Sign-off — user approves the verified artifactreferences/approval-gates.mdGate 3 (interactive)

Gate 2 FAIL loops back to the tagged stage (bounded: 2 cycles, then escalate). Only a PASS advances to Gate 3.

Hard rules

  • No raw markdown as the final external artifact. Render to the format the audience expects (render-guide.md). Markdown is for authoring, not delivery.
  • Voice and anti-slop are mandatory, not optional polish. Every external deliverable goes through voice-master and humanizer before Render.
  • Lead with the strongest material. Decide what_leads at Intake; the verifier checks it.
  • The spec is the state. Carry state in ~/.genesis/sessions/<session_id>/deliverable.json, not in your head — see intake.md for the schema and session-id resolution.
  • Never claim done without a Gate-2 PASS. The Stop-hook enforces this; don't fight it — pass the gate or cancel the deliverable (approval-gates.md).

State machine

drafting → rendered_unverified → verified → shipped (or cancelled). The Stop-hook gate (scripts/hooks/deliverable_gate_guard.py) blocks session-end only in rendered_unverified. Set rendered_unverified right after Render, verified only after a Gate-2 PASS, shipped after Gate 3. If the user abandons it, set cancelled.

Modes

  • Foreground (default): interactive Gates 1 & 3, the Gate-2 policeman, and the Stop-hook backstop. Triggered when the user asks you directly.
  • Autonomous (v2): runs as a Genesis task-executor step (the decomposer assigns this skill when the task plan has a ## Deliverable Frame). Gate 1 is read from that frame — no interview; Gate 2 is still this skill's own check (run in-session — the Task tool isn't available to an executor step); Gate 3 is a Telegram approval handled by the executor (a VERIFYING-phase blocker); escalation is executor-native; the Stop-hook does NOT apply. See references/intake.md → "Autonomous mode" and references/approval-gates.md.

What this does NOT do (yet)

Real Telegram file attachment for autonomous Gate 3 (sends path + summary for now — fast-follow), and the enterprise-ai-skills / hallmark extras. One deliverable at a time. (Real .xlsx with formulas and polished .pptx decks now work when the optional OfficeCLI backend is present — see references/render-guide.md; without it the builder falls back to CSV / pandoc-basic.)

Example

User: "Turn my take-home analysis into the submission packet it should have been." You: Intake (audience = FDE hiring team; format = PDF; what_leads = the pipeline decision + benchmark results; freeze the brief's asks into AC1…ACn) → Gate 1 → Draft → Structure (answer first, methodology to appendix, action-title headings) → voice-master → humanizer → render PDF via pandoc/make-pdf → Gate 2 (fresh reviewer re-reads the original brief, byte-checks the PDF is real, per-criterion verdict) → Gate 3 sign-off → ship.

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.