agentsclimarketplace

Compound

Skill OutlineDriven/odin-claude-plugin/skills/compound

Use when the user says "compound this", after a verified non-trivial outcome that should live in docs/solutions/ or CONCEPTS.md, or when a durable project term needs defining.From its SKILL.md

Install
npx -y skills add OutlineDriven/odin-claude-plugin --skill compound

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • runs commandsInstructs the agent to run 1 command, including `python3 scripts/validate-frontmatter.py <path>`.

SKILL.md

2.6 KB, 573 tokens by cl100k_base, as published. Nobody here has run it

Compound: durable project knowledge

compound writes exactly two surfaces:

  • docs/solutions/<category>/<slug>.md: one learning per solved problem.
  • CONCEPTS.md at the repo root: one definition per project-specific term.

It does not write memory files. User/preference facts are handed to memory-update.

Auto-invoke

Fire automatically after a verified non-trivial outcome:

  • "that worked"
  • "it's fixed"
  • "ship it"
  • "merged"
  • "resolved"
  • "verified"
  • "done"

Auto-fire is permission to evaluate, not permission to fabricate. The reject-by-default gate below still decides whether a doc is warranted.

Reject-by-default gate

A doc is earned, not assumed. All three filters must clear:

  1. Would I forget this? Skip baseline knowledge anyone in this codebase already carries.
  2. Already covered? Update an existing doc rather than duplicate it.
  3. Scope-qualified? A repo-specific quirk qualifies; a general programming truth does not.

If nothing clears the gate, say so in one line and exit. Never fabricate a doc.

Mode routing

Strip mode: tokens from $ARGUMENTS.

ModeTriggerWhat it does
Solution (default)noneDocument one solved problem → docs/solutions/
Vocabularywhen a durable, reusable project term surfacesReconcile CONCEPTS.md: read references/concepts.md
Headlessmode:headlessNon-interactive overlay on whichever mode is active

Writing a solution doc

  1. Read references/schema.yaml and assets/solution-template.md.
  2. Classify the problem: bug or knowledge track.
  3. Check docs/solutions/ for duplicates by grepping frontmatter (title:, tags:, module:, component:).
  4. Write one file: docs/solutions/<category>/<slug>.md.
  5. Run python3 scripts/validate-frontmatter.py <path> until it exits 0.
  6. Read the file back to confirm.

Writing a concept entry

  1. Read references/concepts.md.
  2. Check CONCEPTS.md for an existing definition of the term.
  3. If absent and the term clears the gate, add one entry: a one-sentence local definition and a second paragraph only for non-obvious behavioral rules.
  4. One definition per concept. Refresh on drift, never duplicate.

Commits

One learning per commit. Body carries an ODIN Op: trailer:

Stage only the surfaces compound wrote or edited.

What ships with it: 5 files

26.9 KB alongside SKILL.md, 1 of them executable

references/

scripts/

Keep looking

Skills are one crate of 325,949. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.