Claude Skill

render

Regenerate a render from its recipe in recipes/

LLM Mart · 0 points · 0 views 0 listing impressions 0 install-command copies
Download pche-broken-artist-forge-of-thought-.claude_skills_render-3ef759b.zip · 2 KB

Install

skills CLI npx skills add https://github.com/pche-broken-artist/forge-of-thought/tree/main/.claude/skills/render
Claude Code claude plugin marketplace add https://llmmart.ai/marketplace.json && claude plugin install pche-broken-artist-forge-of-thought@llmmart
Git git clone https://github.com/pche-broken-artist/forge-of-thought.git

The skills CLI installs just this skill, for any of its supported agents. Claude Code installs the whole pche-broken-artist/forge-of-thought collection as a plugin from our marketplace. Git is the plain clone.

Skill manifest

Role: renderer (what a render and a recipe are: CLAUDE.md, Document chain, Renders). A render is regenerated only here, on the principal's explicit command, or by /release (POS.0810); Claude never regenerates on its own judgement — it reports a stale render and offers. Iterating the recipe is working conversation like any other; generating the render is mechanical.

  1. Infer the current project from context ($1, or ask if ambiguous) and read recipes/$0.md. If it does not exist, offer to compose it through /recipe (a genre where one fits, else bare from templates/recipe.md) and stop.
  2. Read the recipe's declared inputs at their current versions. Where the inputs name ledger.md, make it first, python scripts/forge-ledger.py <slug>: the ledger is made at read time (CLAUDE.md, Ledger) and a render reads the file on disk. Every input is read from disk in the run (CLAUDE.md, Isolated reviewers: what holds for every isolated subagent).
  3. Generate in isolation: spawn one subagent (Agent tool, type general-purpose, session model — never a model override) whose prompt names the project, the recipe path, the output path and the render date, and instructs it to read the recipe and its declared inputs, follow the Instructions and Template exactly, and write the file — nothing else. The prompt carries the front-matter block of step 5 verbatim, so the subagent never derives the provenance shape from a previous render. The subagent sees only the recipe and its inputs, never this conversation: a render is derived from the artefacts, not from what was said about them. Its prompt cites what holds for every isolated subagent (CLAUDE.md, Isolated reviewers), no instance fact in the render among it. It must not read the previous render unless the recipe declares it among its inputs (the release-notes genre does, for the released sections), and must not touch the ledger. Prose is hard-wrapped as CLAUDE.md, Document kinds, says for every document. Where the recipe declares languages:, the subagent writes the primary render first and then each translation as a file of its own, scene for scene and line for line, its provenance naming the primary render it was made from; nothing is re-cut. Content only, always Markdown: the render carries no instructions for a conversion — the recipe's ## Format section is never copied into it.
  4. The subagent writes to renders/$0.md, or, where the recipe declares languages, to renders/$0.<lang>.md per language, or to the recipe's output: path if it declares one (e.g. the repository README). An output: path may lie outside the project, in the engine's docs/videos/ (POS.1540): that is the place of the format file of step 7 only, and the Markdown render always stays in the project's renders/. Overwrite freely; history lives in git.
  5. The render opens with YAML front-matter provenance, written by the subagent:
    ---
    project: <slug>
    render: <recipe>
    generated: YYYY-MM-DD
    recipe: recipes/<recipe>.md v<version>
    inputs:
      - <path> v<version>
    ---
    
    A render in one of several languages adds language: <lang>, and a translation translated_from: renders/<recipe>.<primary>.md. Inputs without a version of their own (e.g. CLAUDE.md) are cited by path alone. A render is stale when any version cited in this front-matter differs from the current version of that file, or a cited file no longer exists. This is the one definition; /forge and /check cite it (POS.1070).
  6. Back in the session: scan the written file mechanically for the instance's facts, python scripts/docs-check.py --file <render>; a hit is said aloud and the principal judges it: an instance fact is regenerated out, never edited by hand; a public fixed text of the recipe (POS.0990) or a fact of the engine's own history stands (POS.0950). Verify the file exists and its provenance is correct — the provenance is the row the ledger's Renders table reads, a translation citing the primary render; nobody writes a row — remake the ledger, python scripts/forge-ledger.py <slug>, and report what was rendered from what.
  7. The plain file (CLAUDE.md, Document chain, Renders). Where the recipe carries a ## Format section, run the script of its format — python scripts/md2docx.py or python scripts/md2pptx.py — on the render with --engine pandoc and the reference document and page size the section names; the file lands beside the render. Where the format is video or html-player, run python scripts/md2video.py on each language's render with the format, the pace, the date and, for html-player, the section of the course and that language's name the section names, the file beside the render or at the recipe's output: (docs/videos/, where the script renews the player page). Where pandoc, or Pillow and ffmpeg, is missing, the render stands, the plain file is not made, and that is said aloud. A recipe without the section ends at the Markdown. The designed file is /publish's, never made here; a screenplay has none, and step 8 does not apply to it.
  8. Where published/<recipe>.md exists, say that the published file is now older than its render: the ledger computes its state stale from the dates (SOL.0680), nobody sets it. Remake nothing.
Files (forge-of-thought)
  • SKILL.md 5.5 KB
    ---
    description: Regenerate a render from its recipe in recipes/
    argument-hint: "<recipe> [project-slug]"
    disable-model-invocation: true
    ---
    
    Role: renderer (what a render and a recipe are: CLAUDE.md, Document
    chain, Renders). A render is regenerated only here, on the principal's
    explicit command, or by `/release` (POS.0810); Claude never regenerates
    on its own judgement — it reports a stale render and offers. Iterating
    the recipe is working conversation like any other; generating the render
    is mechanical.
    
    1. Infer the current project from context ($1, or ask if ambiguous) and
       read `recipes/$0.md`. If it does not exist, offer to compose it
       through `/recipe` (a genre where one fits, else bare from
       `templates/recipe.md`) and stop.
    2. Read the recipe's declared inputs at their current versions.
       Where the inputs name `ledger.md`, make it first, `python
       scripts/forge-ledger.py <slug>`: the ledger is made at read time
       (CLAUDE.md, Ledger) and a render reads the file on disk.
       Every input is read from disk in the run (CLAUDE.md, Isolated
       reviewers: what holds for every isolated subagent).
    3. Generate in isolation: spawn one subagent (Agent tool, type
       `general-purpose`, session model — never a model override) whose
       prompt names the project, the recipe path, the output path and the
       render date, and instructs it to read the recipe and its declared
       inputs, follow the Instructions and Template exactly, and write the
       file — nothing else. The prompt carries the front-matter block of
       step 5 verbatim, so the subagent never derives the provenance shape
       from a previous render. The subagent sees only the recipe and its
       inputs, never this conversation: a render is derived from the
       artefacts, not from what was said about them. Its prompt cites what
       holds for every isolated subagent (CLAUDE.md, Isolated reviewers),
       no instance fact in the render among it. It must not read the
       previous render unless the recipe declares it among its inputs
       (the release-notes genre does, for the released sections), and
       must not touch the ledger. Prose is hard-wrapped as CLAUDE.md,
       Document kinds, says for every document. Where the recipe
       declares `languages:`, the subagent writes the primary render
       first and then each translation as a file of its own, scene for
       scene and line for line, its provenance naming the primary
       render it was made from; nothing is re-cut.
       Content only, always Markdown: the render carries no instructions
       for a conversion — the recipe's `## Format` section is never
       copied into it.
    4. The subagent writes to `renders/$0.md`, or, where the recipe
       declares languages, to `renders/$0.<lang>.md` per language, or
       to the recipe's `output:` path if it declares one (e.g. the
       repository README). An `output:` path may lie outside the
       project, in the engine's `docs/videos/` (POS.1540): that is the
       place of the format file of step 7 only, and the Markdown render
       always stays in the project's `renders/`. Overwrite freely;
       history lives in git.
    5. The render opens with YAML front-matter provenance, written by the
       subagent:
       ```yaml
       ---
       project: <slug>
       render: <recipe>
       generated: YYYY-MM-DD
       recipe: recipes/<recipe>.md v<version>
       inputs:
         - <path> v<version>
       ---
       ```
       A render in one of several languages adds `language: <lang>`,
       and a translation `translated_from: renders/<recipe>.<primary>.md`.
       Inputs without a version of their own (e.g. CLAUDE.md) are cited by
       path alone. A render is **stale** when any version cited in this
       front-matter differs from the current version of that file, or a
       cited file no longer exists. This is the one definition; `/forge`
       and `/check` cite it (POS.1070).
    6. Back in the session: scan the written file mechanically for the
       instance's facts, `python scripts/docs-check.py --file <render>`;
       a hit is said aloud and the principal judges it: an instance fact
       is regenerated out, never edited by hand; a public fixed text of
       the recipe (POS.0990) or a fact of the engine's own history stands
       (POS.0950). Verify the file exists and its provenance is
       correct — the provenance is the row the ledger's Renders table
       reads, a translation citing the primary render; nobody writes a
       row — remake the ledger, `python scripts/forge-ledger.py <slug>`,
       and report what was rendered from what.
    7. The plain file (CLAUDE.md, Document chain, Renders). Where the recipe
       carries a `## Format` section, run the script of its format —
       `python scripts/md2docx.py` or `python scripts/md2pptx.py` — on the
       render with `--engine pandoc` and the reference document and page size
       the section names; the file lands beside the render. Where the
       format is `video` or `html-player`, run `python
       scripts/md2video.py` on each language's render with the format,
       the pace, the date and, for `html-player`, the section of the
       course and that language's name the section names, the file beside the
       render or at the recipe's `output:` (`docs/videos/`, where the
       script renews the player page). Where pandoc, or Pillow and
       ffmpeg, is missing, the render stands, the plain file is not
       made, and that is said aloud. A recipe without the section ends
       at the Markdown. The designed file is `/publish`'s, never made
       here; a screenplay has none, and step 8 does not apply to it.
    8. Where `published/<recipe>.md` exists, say that the published file
       is now older than its render: the ledger computes its state
       `stale` from the dates (SOL.0680), nobody sets it. Remake nothing.
    

Comments (0)

Sign in to join the conversation.

No comments yet.

Reviews (0)

No reviews yet.