growth-log
Distil a hard task, a failure or a surprise into a transferable learning entry — root cause, the pattern, the signal that will flag it next time — checked against earlier entries so the same lesson is never logged twice. Use after debugging that took longer than expected, a rollback or rework, a non-obvious decision, or when the user asks what was learned; the entry lands in rinnegan's session memory at capture.
- License 许可
- MIT
- 最近更新
- 2026-09-25
- Stars
- ★ 0
- Forks
- 0
01 前言 Front Matter
- name
- growth-log
- description
- Distil a hard task, a failure or a surprise into a transferable learning entry — root cause, the pattern, the signal that will flag it next time — checked against earlier entries so the same lesson is never logged twice. Use after debugging that took longer than expected, a rollback or rework, a non-obvious decision, or when the user asks what was learned; the entry lands in rinnegan's session memory at capture.
- allowed-tools
- [ "Read", "Grep", "Glob", "Write" ]
03 SKILL.md 正文
Growth Log
Purpose
Produce learning entries that name a pattern rather than narrate an event, so the next session recognises the
situation before repeating the cost. The entry becomes the Learned bullet and a discovery record when
/rinnegan:capture runs; this skill drafts and de-duplicates, and writes a file only on request.
Trigger Conditions
Use it after a task that involved debugging, redoing, rollback or a non-obvious choice, after anything that was
"harder than it should have been", or when the user asks to review what was learned. Skip it for typo fixes,
one-line tweaks and configuration edits that needed no investigation — the test is whether something had to be
figured out. Decisions go to architecture-decision-records; broad recall goes to recalling-context.
Three Rules
| Rule | Meaning |
|---|---|
| Failures outrank successes | One two-hour bug teaches more than three features that worked first time |
| Search before writing | Grep the index and recent Learned sections; a match gains an example, not a twin |
| Every entry transfers | If "next time I see X, I will Y" cannot be written, the pattern is not found yet |
Entry Shape
### <Pattern title — the mechanism, never the incident>
Context: what was attempted, what went wrong or worked unexpectedly well.
Root cause: the mechanism beneath the symptom, reached by asking "why" until the answer generalises.
Pattern: next time <situation>, do <specific action>.
Signal: the observable that says this pattern is active.
Related: ids of earlier entries this extends or contrasts with.
Four to eight sentences. Shorter has not gone deep enough; longer is telling the story instead of the lesson.
Entry types share the shape and shift the weight: a failure leans on root cause; a method that emerged leans on context and pattern; a discovery about a tool or system leans on pattern; a capability change leans on before-versus-after context.
Workflow
- State what happened in one sentence, then ask why repeatedly until the answer is a class of problem, not this instance (a default that changed underneath existing behaviour, not a version number).
- Grep
${CLAUDE_PLUGIN_DATA}/projects/<key>/index.jsonlfor"kind":"discovery"lines and the session notes'Learnedsections using the root-cause keywords. Merge into a match by proposing an added example; otherwise continue. - Draft the entry in the shape above and check it: title names the pattern; a "next time" sentence exists; the signal is specific; root cause is distinct from symptom; related entries are cross-referenced.
- Hand the entry back for the session note. Tag it with two to four kebab-case tags so the index search finds it.
- Only if the user keeps a learning file in the repository (for example
growth-log/<YYYY-MM-DD>.md) and asks for it, append the entry there — the project directory is the only other place this skill writes.
Safety Checks
- Copying a commit message is not an entry; commits say what changed, entries say why it matters.
- One entry per root cause, not per commit; silence is better than a diary line.
- No credentials or private data in an entry, even as an example of what went wrong.
Output Contract
- Entry — the drafted block, with its type and tags
- Duplicate check — the ids searched and whether the entry merged into an existing one
- Next time — the single transferable sentence, restated on its own
04 同仓库更多 Skills
good-skill
Fixture - a skill whose name equals its directory and whose description clears the forty-character floor (N-2).
wrong-name
Too short.
<skill-name>
<What the skill does, then the situations that should trigger it. Concrete and situational, third person, never marketing. Minimum 40 characters — this text is the auto-invocation surface (N-2).>
writing-plans
Turns an agreed design or spec into a step-by-step implementation plan that an engineer with no context could execute. Use when a multi-step change is understood well enough to sequence but no plan exists yet, before touching code. Produces a plan file with exact paths, test-first steps, interfaces between tasks, and no placeholders.
verification-before-completion
Gate that runs before any claim that work is done, fixed, passing, or shipped. Identifies the command that would prove the claim, runs it fresh, reads the whole output, and only then reports — with the evidence attached and any gap stated. Use before committing, opening a pull request, marking a task complete, or telling the user something works.
using-skills
Session discipline that fires before the first action of any task. Checks whether an installed skill or command already covers the job, keeps every action inside the requested scope, and reports verification as evidence rather than confidence. Use at the start of a conversation, before answering a question, before exploring a codebase, and whenever a task changes shape mid-session.