| name | ai-discover |
| description | Parallel discovery of performance hotspots (perf track) and failure surfaces (bug track) for the auto-improvement loop. Fans out one subagent per hot-path area or failure surface; each returns ONE concrete, behavior-preserving fix candidate (perf) or a reproducing test plus fix (bug). Discovery only — no code changes and no verification happen here. |
| always | false |
| triggers | auto-improvement discovery, find hotspots, profile discovery, discover defects, improvement candidates |
ai-discover — the discovery step of the auto-improvement loop
This skill drives Phase A (discovery) of an auto-improvement cycle. It is
discovery only: it never applies a change, never runs the keep-or-revert A/B,
and never decides whether a finding is kept or drafted as a pull request. Those
are the spine's deterministic Python gate / keeper / pipeline, which no model
can argue past. That separation is the point — the measurement is the product.
Perf track
- Fan out one subagent per candidate hot-path area.
- Each runs the profiler against a realistic workload and finds the single
largest behavior-preserving win in its area.
- Each returns exactly ONE candidate:
{target locus, signature, hypothesis, expected stage win, scenario}.
- Write all candidates to the discovery artifact the spine reads. The spine
dedups by content fingerprint, implements, and verifies with a serial pinned
A/B measurement. Discovery does none of that.
Bug track (kind="bug")
- Fan out one subagent per failure surface — a risky or known-fragile path.
- Each writes a minimal deterministic reproducing test that FAILS on the
base commit (RED) and a fix that makes it PASS (GREEN) without regressing the
rest of the suite.
- Each returns
{target locus, reproducing test id + path, fix diff, blast radius, severity note}. The spine's RED → GREEN → STAYGREEN gate is the
verdict; there is no A/B or noise band for bug findings.
Reporting a candidate honestly
State the expected win as a hypothesis, not a result. The spine measures it;
if your estimate was wrong the candidate is reverted and that is a normal,
useful outcome. A guess presented as a measurement is the failure mode this
whole app exists to prevent.
What this skill never does
- Never edits the ruler, the measurement harness, the tests-of-record, or
anything outside the active target profile's edit allowlist. Those paths
are mechanically rejected, so an edit there wastes the whole cycle.
- Never publishes or merges a pull request. Survivors are drafted as GitHub
draft PRs by the spine, and a human publishes them.
- Never fabricates a measured number. A fabricated win is the worst possible
reward-hack, because it corrupts the record the loop reasons from.