Motivation
From the personal PR-reviewer evaluation (kept outside this repo) and
#1056 (local-strict-review skill
proposal). agent-conduct already fires at the moment an agent is "about to claim work is done,
verified, green, or fixed" — that's the exact moment the review-volume catch-22 happens: a local
agent judges a unit of work ready, opens the PR, and only then discovers what a stronger-model
adversarial pass would have caught for free, locally, before ever pushing.
Proposal
Extend agent-conduct's existing "about to claim work is done" trigger so that, specifically for
PR-bound changes, it points at running local-strict-review (once #1056 lands) before the claim
is made — the same way that trigger moment should already be routing to other verification
discipline. Small, cheap addition to an existing trigger; no new trigger moment needed.
Scope note
This issue is just the agent-conduct wiring. The skill itself is #1056; the broader PR-flow
chaining (lint -> strict-review -> future CodeRabbit CLI) is a separate issue for
pr-review-conduct/drive-pr. This one is specifically about the "declaring done" moment an
agent hits mid-task, which is earlier and more general than the PR-open moment those other two
skills govern.
Motivation
From the personal PR-reviewer evaluation (kept outside this repo) and
#1056 (
local-strict-reviewskillproposal).
agent-conductalready fires at the moment an agent is "about to claim work is done,verified, green, or fixed" — that's the exact moment the review-volume catch-22 happens: a local
agent judges a unit of work ready, opens the PR, and only then discovers what a stronger-model
adversarial pass would have caught for free, locally, before ever pushing.
Proposal
Extend
agent-conduct's existing "about to claim work is done" trigger so that, specifically forPR-bound changes, it points at running
local-strict-review(once #1056 lands) before the claimis made — the same way that trigger moment should already be routing to other verification
discipline. Small, cheap addition to an existing trigger; no new trigger moment needed.
Scope note
This issue is just the
agent-conductwiring. The skill itself is #1056; the broader PR-flowchaining (lint -> strict-review -> future CodeRabbit CLI) is a separate issue for
pr-review-conduct/drive-pr. This one is specifically about the "declaring done" moment anagent hits mid-task, which is earlier and more general than the PR-open moment those other two
skills govern.