oliverweissl/context-garden / skills/seedbank
seedbank
VerifiedMaintain 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.
| File | Kind | Size |
|---|---|---|
| .gitignore | other | 40 B |
| SKILL.md | skill md | 2.7 KB |
| agents/gardener.md | markdown | 1.6 KB |
| bin/seedbankexecutable | script | 1.2 KB |
| references/guide.md | markdown | 3.3 KB |
| references/schema.md | markdown | 5.4 KB |
| references/scoring.md | markdown | 2.1 KB |
| scripts/cg_fingerprint.py | script | 3.1 KB |
| scripts/cg_mode.py | script | 5.4 KB |
| scripts/sb_compile.py | script | 5.6 KB |
| scripts/sb_conflicts.py | script | 2.3 KB |
| scripts/sb_hook.py | script | 13.5 KB |
| scripts/sb_scoring.py | script | 3.4 KB |
| scripts/sb_store.py | script | 9.3 KB |
| scripts/seedbank.pyexecutable | script | 43.0 KB |
| tests/fixtures/mini_repo/CMakeLists.txt | markdown | 94 B |
| tests/fixtures/mini_repo/README.md | markdown | 126 B |
| tests/fixtures/mini_repo/src/generated/bindings.cpp | other | 92 B |
| tests/fixtures/mini_repo/tools/codegen.py | script | 57 B |
| tests/smoke_test.shexecutable | script | 29.7 KB |
references/scoring.md
View on GitHub# Promotion scoring
```
value = ((distinct_sessions - 1) * avg_retrieval_cost + total_failure_cost)
* stability * decay / max(1, persistent_token_cost)
decay = 0.5 ** (days_since_last_access / half_life_days) # 1.0 for --critical facts
```
Implemented in `scripts/sb_scoring.py`. Heuristic ranking, not measured cost.
- **`distinct_sessions`** — distinct session ids that observed the key:
`--session`, else `$CLAUDE_CODE_SESSION_ID` (or `$CLAUDE_SESSION_ID`),
else today's date + terminal session id; the hook uses the hook JSON's
`session_id`. Only the last 32 ids are kept, so an id that fell out of
that window counts again. Repeat reads within one session count once.
- **`distinct_sessions - 1`** — only rediscoveries count: a single read of a
10k-token file scores 0.
- **`avg_retrieval_cost`** — `total_retrieval_cost / access_count`:
`read` = file chars/4, `search` = `--cost` or 30, `run` = captured output
chars/4, `mistake`/`fact` = 0.
- **`total_failure_cost`** — summed per occurrence: `mistake` 300
(`--failure-cost`), failed `run` 150, `fact` `fact_failure_cost` (60,
`config.json`), so a never-rediscovered invariant still scores > 0.
- **`stability`** — `1 / (1 + hash_changes)` of the key's sources.
- **`decay`** — `half_life_days` (default 30) since last access. For
promoted facts (hot-budget eviction), promote and `invalidate --confirm`
count as an access.
- **`persistent_token_cost`** — `chars(representation) / 4`; 20 for a
candidate without a representation.
## Threshold
`candidates` lists keys with `value >= promote_threshold` (`config.json`,
default 1.0). At the default: a ~30-token file rediscovered in 2 sessions
≈ 1.6; a compact logged mistake ≫ 1; a declared fact ≈ `60 / tokens`,
dropping below 1 after about two half-lives without access.
## Worked example (from `tests/fixtures/mini_repo`)
- `README.md` (~32 tok) read in 5 sessions: `(4 * 32 + 0) / 20 = 6.4`.
`big.txt` (10k tok) read once: `(0 * 10000 + 0) / 20 = 0`.
- Mistake logged 10x (~14 tok): `3000 / 14 ≈ 214`; logged once: `≈ 21`.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| 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: 472k | Updated | Claude CodeCodex | Unverified | Install |
| 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: 322k | Updated | Claude CodeCodex | Verified | Install |
| dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| 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 available | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |