Skip to content

oliverweissl/context-garden / skills/seedbank

seedbank

Verified

Maintain a small, generated AGENTS.md of repository facts (build commands, conventions, invariants, past mistakes) that were rediscovered often or were costly to get wrong. Use at task start to check cached facts, and at task end to log and promote what you had to dig for.

seedbank

!"${CLAUDE_SKILL_DIR}/bin/seedbank" mode --banner

Agents that do not expand the line above (e.g. Codex): run <this-skill-dir>/bin/seedbank mode --banner first and follow it.

Offline fact store + profiler. It counts rediscoveries and mistake cost; you supply the fact text and decide what to promote. Nothing is promoted just because it was seen once; promotion is always a deliberate promote call. Formulas: references/scoring.md.

seedbank is not on PATH: always run it as <this-skill-dir>/bin/seedbank. State lives in <git repo root>/.seedbank/.

  1. Start: read AGENTS.md. If it answers your question, skip re-discovery.
  2. While working, log only "how does this repo work" discoveries, not every file you touch (with the plugin hook, Read/Grep/Glob and read-only Bash such as cat/sed/grep/rg/find are logged automatically):
    bin/seedbank observe run --scope <topic> -- <command...>
    bin/seedbank observe mistake "<rule you broke>" --scope invariants [--source <file>]
    bin/seedbank observe fact "<invariant>" --scope invariants
    
  3. End: bin/seedbank candidates, then
    bin/seedbank promote <key> --representation "<compact text>" --tier hot [--critical]
    bin/seedbank promote <key> --tier warm --scope <topic>
    bin/seedbank compile
    
    A correctness/safety invariant is always worth promoting: use --critical so it's protected from budget-based eviction. Promote refuses a fact contradicting one in the same scope; re-run with --replace <id> or --force. compile rewrites only the region between <!-- seedbank:begin -->/<!-- seedbank:end --> and excludes stale facts (source changed since promotion) instead of serving them as still true; --critical facts stay, marked "(source changed — verify)". With the plugin, a Stop hook reminds you once per session when this session produced candidates; hand the curation to the gardener agent to keep it out of your context, or do step 3 yourself.
  4. Existing hand-written AGENTS.md/CLAUDE.md: bin/seedbank import <path>.
  5. Sharing facts: bin/seedbank export --out f.json; the recipient runs bin/seedbank import f.json (refused unless the repo fingerprint matches).

Hook setup, tiers, what to commit: references/guide.md. Full CLI: references/schema.md.

reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis.Stars: 471kUpdated Claude CodeCodexVerifiedInstall
diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness.Stars: 322kUpdated Claude CodeCodexVerifiedInstall
dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependenciesStars: 322kUpdated Claude CodeCodexVerifiedInstall
executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is availableStars: 322kUpdated Claude CodeCodexVerifiedInstall