{
  "markdown": "# skills\n\nCanonical agent skills repo. Single source of truth — consumer repos install copies via `npx skills`, never edit skills locally.\n\n## Layout\n\n```text\noriginal/   # own skills, authored here\nderived/    # condensed vendor derivations, pinned via metadata.derived-from\ndocs/       # architecture + glossary (ubiquitous language)\ntests/      # pressure scenarios and fixtures, never shipped\n```\n\n## Installing\n\nFrom a consumer repo:\n\n```sh\n# list available skills\nnpx skills add frankify-app/skills --list\n\n# install selected skills (project scope, Claude Code)\nnpx skills add frankify-app/skills --skill documenting-decisions -a claude-code -y\n\n\n# restore on a fresh clone\nnpx skills experimental_install\n\n# update\nnpx skills update -p -y\n```\n\n## Glossary\n\nUbiquitous language: `docs/glossary/`, one file per term, resolve via `uvx disambiguate <term>`:\n\n- [Skills](docs/glossary/skills.md)\n- [Grilling Session](docs/glossary/grilling-session.md)\n\n## Rules\n\n- Edit skills here only. Local copies in consumers get overwritten on update.\n- As short as possible — prefer caveman mode (`derived/caveman`) for skills, ADRs, tickets, docs. Hard limit: precision and understandability must not suffer.\n- Skills stay tool-agnostic: no consumer-repo tooling (tracker CLIs etc.) hardcoded — defer to consumer's AGENTS.md conventions.\n- Descriptions state triggers only (\"Use when ...\"), never workflow.\n- Test artifacts live in `tests/`, never inside skill folders. Fixture manifests are named `SKILL.fixture.md` so discovery ignores them.\n\n## Adding a skill\n\n1. `original/<name>/SKILL.md` — frontmatter `name` (matching the folder, letters/numbers/hyphens only) + `description`.\n2. Keep SKILL.md short; split heavy reference into separate files inside the skill folder.\n3. For discipline-enforcing skills: baseline-test against an agent without the skill first; scenarios go in `tests/<name>/`.\n4. Consumers pick it up via `npx skills add frankify-app/skills --skill <name>`.\n\n## Adding a derived skill\n\n1. Copy the original skill folder into `derived/<skill-name>/`\n2. Commit the skill in original condition, verbatim — its own commit, e.g. `chore(<skill-name>): copy upstream skill verbatim`, so the derivation stays diffable against upstream\n3. Add `metadata.derived-from: [info] <full commit URL>` from the original repo into frontmatter (To update: diff pin against upstream HEAD, fold changes into derivation, bump pin)\n4. Edit the skill and commit the derived version separately\n",
  "bytes": 2484,
  "sha": "8978cdd5e210cca417bc47f0b1c51c6f81b80342843cd2b24df38663a7293e72",
  "repo_slug": "pandoscope/skills",
  "fonte": "repo",
  "truncated": false,
  "api": "https://agentalog.com/api/listings/skl_pandoscope_skills_caveman_de6fb03d/readme"
}