Log
Skill RockyHong/super-bootstrap/plugins/super-bootstrap/skills/log
Claude Code plugin: one command scaffolds CLAUDE.md, picks skills/MCPs for your stack, and adds a route-aware workflow with doc-sync commit gate.
npx -y skills add RockyHong/super-bootstrap --skill logAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
One thing to look at
- 2 stars2 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
What its author says it does
Copied from the file, not written here
Capture front door. Logs 1..N observations as card files at docs/work/{ID}.md (BUG / DEBT / GAP), classified + deduped on Sonnet. A duplicate carrying a new fact lands as an Amendment block on the owning card (amended branch). Use when the user says "log this", "track that", "note this down", or types `/super-bootstrap:log <observation>` — and when Claude needs to file its own findings (out-of-scope findings from a review, audit, or returned subagent report). Dispatches the `log` subagent so classify + write run off the gateway model. Captures raw — the real/worth/now call is triage's at `/super-bootstrap:todo` pickup. Does NOT triage (that is the `/super-bootstrap:todo` triage lane). Feature ideas log as GAP.
SKILL.md
4.0 KB, as published. Nobody here has run it
Log — Capture Front Door
Muscle-memory capture. Takes whatever the caller hands it — one observation or a batch — classifies each into BUG / DEBT / GAP, dedups against open cards, and writes each as a card file at docs/work/{ID}.md. A duplicate carrying a genuinely new fact lands as an Amendment block appended to the owning card (amended output branch). Capture is unconditional (bugs, debt, design gaps, unverified ideas all land); the real/worth/now call runs at /super-bootstrap:todo triage on pickup. The thinking runs in the log subagent (agents/log.md, model: sonnet); this skill is the dispatch shell.
All new cards route through here — user-initiated and Claude-initiated captures alike. Sanctioned alternative: hand-copy docs/work/TEMPLATE.md and bump the ID high-water line in docs/work/README.md in the same write.
When it fires
- User — explicit
/super-bootstrap:log <observation>, or natural-language "log this / track that / note this down". - Claude — its own captures: a bug spotted mid-task, the out-of-scope findings a review or returned subagent surfaced. Batch them into one invocation — never one dispatch per finding.
A card is owed only for deferred or dropped work — work that exits the current flow incomplete. Work completed in-flow, whoever directed it, carries no card debt. The trigger is completion-state (observable), not worth.
Out of lane: triage (root-cause investigation → the /super-bootstrap:todo triage lane). Feature ideas and unverified hunches are in lane — they log as GAP. This skill creates new cards and new-fact Amendments only.
Arguments
The argument is the raw observation(s). Free-form. May be one item or many (a list, a pasted findings block). The skill does not parse buckets from the arg — the subagent classifies.
| Invocation | Behavior |
|---|---|
/super-bootstrap:log <text> | Dispatch the log subagent with the text as entries. |
/super-bootstrap:log (bare) | Ask the caller what to capture (one line), then dispatch. |
Execution
- Gather the entries — the user's text and/or the findings block the gateway is holding. Keep each entry's context (where it came from, any source file/line) so the subagent can dedup + write a faithful row.
- Dispatch:
Agenttool,subagent_type: "log", prompt = the entries (1..N) + any source context, verbatim, + today's date. Do not pre-classify, do not pre-judge buckets — that is the subagent's job, and pre-judging feeds it bias. - Relay the subagent's return to the caller verbatim. Return shape:
agents/log.md§ Output contract. - Spot-check: sample one written row against the entry it captures; a confirmed miss →
/super-bootstrap:log(tier re-pinning evidence).
Rules
- Route input to the subagent; the subagent classifies. The category + dedup decisions live in the
logsubagent. This shell routes input and relays output. - Batch over loop. Many findings → one dispatch with all entries. Per-finding dispatch is the anti-pattern this skill exists to avoid.
- Relay questions, never auto-answer. Thin context is the subagent's signal to ask; the caller resolves it. Silent guessing files wrong rows the caller then trusts.