build-pack
Skill3 starsChanged 24 days agoMITView on GitHub
Builds one GTM skill pack through a knowledge card, a design card, and a pack gate. Use when starting a pack, before the pack is written, or when a scorecard, license, secret, README, or install line is about to ship.
1 other project uses a version of this.
Adopted by
Most recently taken up by
- yantrikos/yantrikdbtheir own version, first seen Jul 29, 2026
Other versions1 project uses a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 1
- yantrikos/yantrikdb★ 73same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository7 other files it gives its agents.
- AGENTS.mdAGENTS.md
- skills/cold-email/AGENTS.mdAGENTS.md
- skills/evp/AGENTS.mdAGENTS.md
- skills/founder-brand/AGENTS.mdAGENTS.md
- skills/pricing/AGENTS.mdAGENTS.md
- skills/psp/AGENTS.mdAGENTS.md
- company-context-packSkill
The file
Files
Build a skill pack
One pack finishes one job. The script is the gate. Cards are paths. They do not have to be committed.
Python 3.10 or newer, stdlib only. No install. Run the commands from this skill's base directory, or put that directory in front of scripts/check_pack.py. The <dir> and <card> paths point at the pack you are building.
python3 scripts/check_pack.py knowledge <card>
python3 scripts/check_pack.py design <card>
python3 scripts/check_pack.py pack <dir> --public
python3 scripts/check_pack.py pack <dir> --private
python3 scripts/check_pack.py all <dir> --public --knowledge <card> --design <card>Success prints knowledge ok, design ok, or pack ok. A miss prints Refusal: and exits 1.
Checklist
Copy this list and tick it in order.
- 1. Knowledge. Fill the card. Run the knowledge command. Read knowledge.md.
- 2. Design. Fill the card. Run the design command. Read design.md.
- 3. Implementation. Write the pack.
nameis lowercase letters, numbers, and single hyphens, at most 64 characters, and never holdsclaudeoranthropic. The description says what the pack does and when to use it, in under 1024 characters, with no angle-bracket tags.modelsstays empty until a named model has run it. For a Claude pack, probe a fast model, a mid model, and a large model: enough guidance, clear and efficient, and no extra explanation. A fast model that skips a step needs a clearer line or a script. A large model that does worse with the skill than without it needs fewer lines. If the only way to hold the line is all caps, make it a script. The body stays under 500 lines. Every SKILL.md loads: it lives inskills/<name>/, or plugin.json lists its folder underskills. An agent inagents/has frontmatter and takestools, notallowed-tools. A frontmatter key holds a value or a list, never both. Call a bundled script by${CLAUDE_PLUGIN_ROOT}or${CLAUDE_SKILL_DIR}, never by a path that depends on the working directory. A reference over 100 lines has a Contents list inside the first 100 lines, and the bullets match the later headings. Link every reference from this file, one level deep. Run a script. Do not paste the script into the reply. A third-party import keeps its install line in the five lines above it. One pack, one skill: other jobs are mode files underskills/<name>/modes/that the skill reads on demand, so the always-on description stays under about 400 tokens;argument-hintlists the modes. Drafts the pack asks for go ingtm/<pack>.jsonat the project root, never a shared name. Scorers take--fileand--stdin, print text by default and one JSON object with--json, exit 0 pass, 1 refused, 2 bad input, and end with aNext:line. Ship 4 to 6 cases underevals/that check the skill fires. - 4. Security. Write
SECURITY.mdat the pack root: what runs locally, what it reads and writes, the network it opens (or "None"), no telemetry, no credentials, and where to report. A public pack without it is refused. Run the pack command with--publicor--private. Read pack-checks.md. - 5. UX. The README is the landing page. Read ux.md.
- 6. Scorecard. Run the scorer, fix, run the scorer, and check again until it passes. A miss the scorer does not know is a proposed rule. Leave it for a person to approve.
- 7. Release. Open a pull request from a worktree of origin/main. Leave it unmerged. Do not upload a zip by hand.
Go back to step 3 if step 4 or step 6 fails. Go back to step 2 if step 5 fails.
Passing shapes: examples/knowledge.json and examples/design.json.
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.

