PocketForge
AGENTS.md0 starsChanged 7 days agoMITView on GitHub
This file is read by Codex, Google Antigravity, Cursor, GitHub Copilot, Windsurf, Gemini CLI (via GEMINI.md), Amp, Jules and other agents that support AGENTS.md. Claude Code users can install the plugin instead (see README). PocketForge turns a skill or niche into a sourced, illustrated ebook. The pipeline is plain files plus Node/Python scripts, so any agent that can read files and run shell commands can run it. Web research needs a Firecrawl MCP server (or another search/scrape tool you configure;…
- Installs packages
Where files are headingNew AGENTS.md files have more rules about keeping secrets out of the code, and fewer about code style.
Of the AGENTS.md files we read, the 32,671 first seen in the last six months, against the older ones: the kind of rule that gained most share, and the kind that lost most. About files like this in general, not this one, and an estimate, since a kind is a pattern in the words.
In this repository7 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- templates/CLAUDE.mdCLAUDE.md
- .github/copilot-instructions.mdCopilot instructions
- .cursor/rules/pocketforge.mdcCursor rule
- anti-slop-writingSkill
- ebook-structureSkill
- mascot-rigSkill
The file
We split this AGENTS.md into the rules it gives an agent
3 rules. About 0 are common, 1 is distinctive, and 2 are 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.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 35
Never invent stats, quotes, testimonials, names or credentials.
- Very specific: line 36
After writing or editing projects/<slug>/manuscript/chNN.md, run node scripts/gates/run-all.mjs <file>.
- Distinctive: line 39
Ask before global installs, deleting files, or adding MCP servers.
≈ “Common” is an estimate: at least 1 in 5 of the 41,581 AGENTS.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
PocketForge — instructions for any AI coding agent
This file is read by Codex, Google Antigravity, Cursor, GitHub Copilot, Windsurf, Gemini CLI (via GEMINI.md), Amp, Jules and other agents that support AGENTS.md. Claude Code users can install the plugin instead (see README).
PocketForge turns a skill or niche into a sourced, illustrated ebook. The pipeline is plain files plus Node/Python scripts, so any agent that can read files and run shell commands can run it.
Read first
templates/CLAUDE.md: the full rule book (phases, hard stops, truth rules, writing rules, gates, images, mascot, folder layout). Treat it as binding, whatever its filename says.skills/*/SKILL.md: load the relevant skill before its task:anti-slop-writing: before writing or critiquing any chapter, blurb or listing.ebook-structure: before outlining, writing or laying out.mascot-rig: before building, posing or animating the mascot.
agents/*.md: role playbooks (researcher, gap-miner, pocket-scorer, benchmarker, outliner, writer, critic, character-designer, animator, designer, layout). If your tool has no subagents, play each role yourself in order, and start a fresh context or chat for the critic role so the writer never grades its own chapter.commands/*.md: the entry points.forge.mdis the full pipeline;pockets.md,book.mdandanimate.mdare partial runs. Follow them as checklists.
Setup (once per workspace)
npm install # playwright, resvg, ffmpeg-static, hyperframes (local only)
pip install pillow pypdf pymupdf # EPUB build and PDF checks
npm test # gate tests must passWeb research needs a Firecrawl MCP server (or another search/scrape tool you configure; every research file must still start with Source: <URL> | Fetched: <YYYY-MM-DD>). Canva and Higgsfield are optional.
Phases (stop where marked and wait for the user)
0 skill check · 1 research · 2 gap mining · 3 pockets (STOP: user picks one) · 4 transformation map · 5 benchmark · 6 outline (STOP: approve title + TOC) · 7 manuscript · 8 character sheet (STOP: approve mascot) · 9 poses + scenes · 10 animations · 11 brand + cover (STOP: pick a cover) · 12 layout · 13 export + extras.
Record each approval in projects/<slug>/approvals.md. Print ✅ Phase N: <result> after each phase.
Non-negotiables
- Never invent stats, quotes, testimonials, names or credentials. Every stat gets a
[^n]footnote that resolves to a URL on line 1 of a file inprojects/<slug>/research/. - After writing or editing
projects/<slug>/manuscript/chNN.md, runnode scripts/gates/run-all.mjs <file>. A chapter is done only when all gates pass and a fresh critic passes it. (Claude Code runs this automatically through a hook; other agents must run it by hand.) - Hypothetical math goes in
:::exampleblocks; worksheets in:::worksheetblocks. - Health, legal and finance books get a disclaimer page.
- Ask before global installs, deleting files, or adding MCP servers.
Build commands (per project)
node scripts/layout/compose-cover.mjs projects/<slug> # needs assets/higgsfield/cover.png
node scripts/layout/html-to-pdf.mjs projects/<slug> # output/ebook.pdf + book.html
NO_COVER=1 PAGE_W_IN=6 PDF_OUT=paperback-interior.pdf node scripts/layout/html-to-pdf.mjs projects/<slug>
node scripts/layout/paperback-cover.mjs projects/<slug> <pages> [color|white]
python scripts/layout/build-epub.py projects/<slug>
node scripts/extras/worksheet-pack.mjs projects/<slug>
node scripts/extras/build-extras.mjs projects/<slug> # mockup + listing images
node scripts/extras/hf-art.mjs projects/<slug> # optional Higgsfield artBook metadata (title, author, mascot, blurb, listing copy, art prompts) lives in projects/<slug>/book.json; copy templates/project/book.example.json to start.
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.

