reconcile: review-feedback classifier — normalize upstream events into …

[warren-run-bot]

Summary

feat(campaign-controller): review-feedback classifier via profile-declared bot grammar (warren-2ec3)

Run

• Warren run: run_h9cmhtxh2twm
• Agent: pi
• Cost: $0.672 (62.7k in / 23.0k out / 3.2M cache-r)

Seeds

• warren-2ec3 — reconcile: review-feedback classifier — normalize upstream events into actionable categories via profile-declared bot grammar (untrusted-input discipline)

Commits (2)

• 7bb2a71 feat(campaign-controller): review-feedback classifier via profile-declared bot grammar (warren-2ec3)
• c7ff72f chore: close warren-2ec3, record classifier pattern (warren-2ec3)

Files changed

.../profiles/openclaw.repository-policy.json       |  10 -
 .../src/openclaw-profile.test.ts                   |  30 +-
 .../src/reconcile/bot-grammar.ts                   | 182 +++++++++
 .../src/reconcile/classifier.test.ts               | 432 +++++++++++++++++++++
 .../src/reconcile/classifier.ts                    | 233 +++++++++++
 .../src/reconcile/reconciler.ts                    |  44 +++
 .../src/repository-policy.test.ts                  |  84 +---
 .../campaign-controller/src/repository-policy.ts   |  66 ----
 extensions/campaign-controller/src/store/events.ts |  88 ++++-
 extensions/campaign-controller/src/store/schema.ts |  18 +
 .../src/store/state-store.test.ts                  |   4 +-
 extensions/campaign-controller/src/store/types.ts  |  12 +
 .../campaign-controller/src/tick/tick.test.ts      |  32 --
 extensions/campaign-controller/src/tick/tick.ts    |  27 +-
 14 files changed, 1014 insertions(+), 248 deletions(-)

Prompt

Show prompt

Work seeds issue warren-2ec3. First run `sd show warren-2ec3 --json` from the repo root — the issue body is the full spec; this summary is only a digest.

Task digest: build the review-feedback classifier in the campaign-controller reconcile path: normalize upstream PR events (review-bot comments, reviews, check results) into actionable categories via profile-declared bot grammar (bot logins, marker formats, re-review commands as data). Untrusted-input discipline is mandatory: pass through only structured fields (check names, file paths, finding titles) with provenance marked untrusted — never raw comment bodies as instructions. Classification only; do not build the follow-up coordinator that consumes it (sibling issue warren-0ad3).

Context: this issue is one step of plan pl-096b (upstream contribution loop v1, repo-agnostic phase 3) for the campaign-controller. The extension lives in extensions/campaign-controller/ — a standalone package with its own package.json, lockfile, and tests. For extension work, run `cd extensions/campaign-controller && bun install && bun test` and keep the extension green. Extensions never import src/ or scripts/ (check:layers enforces this).

IMPORTANT — parallel session: several sibling issues from this plan are being worked simultaneously by other agents. Keep your diff strictly scoped to what this issue requires; do not refactor shared files beyond your scope, or the PRs will conflict.

Quality gates are terminal: `bun run check:all` from the repo root must be green before you commit and report done. If you touched extensions/campaign-controller, its own `bun test` must also pass.

When the work is complete and gates are green: close the issue with `sd close <id> --reason "<evidence-citing reason>"`, then commit everything (including .seeds/ changes).

---

🤖 Opened by warren run run_h9cmhtxh2twm