feat(review): hunter-over-engineering — 7º track repo-wide de sobre-ingeniería - #4
Conversation
- hunter-over-engineering work in progress (interrupted by model switch)
|
Hola, @ixjosemi: Cuando aceptes la invitación, te añado como requested reviewer. Un abrazo |
|
@ixjosemi te comparto esto por si te apetece echarle un vistazo, sin ningún compromiso — solo era para compartirlo 😄 Es un fork de Convoy en mi GitHub (dagocareer/convoy). Me pareció una herramienta muy útil para el día a día y le he añadido cuatro features que me faltaban. Lo curioso es que toda la implementación se ha hecho usando el propio Convoy (dogfooding): cada feature se escribió con su pipeline Un resumen de cada una, en una frase:
Están todas abiertas y mergeables por si te apetece revisarlas, pero vamos, sin compromiso. |
feat(review): hunter-over-engineering — 7º track repo-wide de sobre-ingeniería
Adds a repo-wide over-engineering audit as the seventh specialty track of the
hunterandhunter-maxpipelines. Unlike the diff-scopedover-engineering-auditor(PR 2), this track scans the whole repository whenprd.mdnames no narrower scope, hunts unearned abstraction / indirection / configurability / speculative generality / removable code, ranks findings by largest cut first, and reports the net lines and dependencies removable. Inspires from the Ponytailponytail-auditskill, but is implemented natively: same hunter fan-out, consensus, statistics, and fingerprinting as the other six tracks.This is PR 4 of a 4-PR program inspired by Ponytail — Convoy dogfooding itself through its own implement pipeline. Based on
main; independent of PRs 1–3.What changed
prompts/hunter-over-engineering.md(new): repo-wide over-engineering audit prompt in the standardhunter-*shape (specialty, scope, limitations, findings with severity/confidence/location/evidence/trigger/fingerprint), with the N1–N7/V1–V7 taxonomy and the seven tags (yagni,premature-abstraction,configurability,indirection,complexity,over-processing,scope-creep) defined inline so it is self-contained onmain. Findings are ranked by biggest cut first; the report closes with the net lines and dependencies removable, and says so plainly when there is nothing to cut.src/pipeline.ts: adds"over-engineering"tohunterTracks, registers thehunter-over-engineeringagent (readOnly: true,temperature: 0.1), and wires it into thehunterfan-out as[fallbackModel, glmModel]— it flows intohunter-maxautomatically throughhunterMaxTracks().src/built-in-prompts.ts: registers the new prompt (sync test covers it).prompts/hunter-report.md/prompts/hunter-max-report.md/ README.md /src/config.ts: expected coverage updated from six to seven specialties (30 → 35 audits in hunter-max).Verification
bun run typecheck— passbun test— 755 pass, 0 failbun run build— passOut of scope (follow-ups)
hunter/hunter-maxspecialty model roster unchanged for the existing six tracks; no new pipeline;review/refine/review-ccuntouched.