Convergence controls: max rounds / diminishing-returns stop #84

New issue

Closed

opened 2026-10-15 19:45:37 +00:00 by agent-pi · 1 comment

agent-pi commented

2026-10-15 19:45:37 +00:00

Collaborator

Convergence controls would bound this: e.g. a configurable cap on review rounds per PR, and/or an exit rule where a wave containing only low-severity findings is summarised in a single comment instead of posted inline (signalling diminishing returns).

On an active PR, every fix-push triggers a fresh review wave, and there is no natural stopping point: #75 ran five waves with increasingly marginal and repetitive findings, and would have continued indefinitely. Convergence controls would bound this: e.g. a configurable cap on review rounds per PR, and/or an exit rule where a wave containing only low-severity findings is summarised in a single comment instead of posted inline (signalling diminishing returns).

agent-pi added this to the Review quality & convergence milestone

2026-10-15 19:45:37 +00:00

agent-pi added the

Kind/Enhancement

Priority

Low

labels

2026-10-15 19:45:37 +00:00

agent-pi referenced this issue

2026-10-16 23:38:44 +00:00

action.yml missing wave-1 inputs (verify, min_severity) #107

agent-pi referenced this issue

2026-10-16 23:39:09 +00:00

Conversational re-reviews: feed the previous review thread back #79

agent-pi commented

2026-10-16 23:39:10 +00:00

Author

Collaborator

Wave 2 plan (last of the wave, after #79/#85 land). Both controls from the issue body, sharing the structured reviews fetch from #79.

Plan of attack

Branch: feat/issue-84-convergence-controls
Hard cap: WUMING_MAX_ROUNDS (default 0 = unlimited). Round = count of prior ## WuMing reviews on the PR + 1. At/over the cap: no agent calls; post one plain PR comment ("cap reached, skipping further waves") once — deduped via the existing marker scan — and exit 0
Soft diminishing-returns stop: when a wave produces findings and all are below high severity and a prior review exists, post them as a single plain PR comment instead of an inline review (a wave of LOW nits stops pinning the diff)
Both fail open: if prior reviews cannot be fetched, behave as today
Config + action.yml inputs in the same PR; README, CLAUDE.md, Release-Notes.md; tryke tests (round counting, cap behaviour, low-only routing, fail-open)

Refs the wave plan on #79.

Wave 2 plan (last of the wave, after #79/#85 land). Both controls from the issue body, sharing the structured reviews fetch from #79. **Plan of attack** - Branch: `feat/issue-84-convergence-controls` - **Hard cap**: `WUMING_MAX_ROUNDS` (default `0` = unlimited). Round = count of prior `## WuMing` reviews on the PR + 1. At/over the cap: no agent calls; post one plain PR comment ("cap reached, skipping further waves") once — deduped via the existing marker scan — and exit 0 - **Soft diminishing-returns stop**: when a wave produces findings and *all* are below `high` severity *and* a prior review exists, post them as a single plain PR comment instead of an inline review (a wave of LOW nits stops pinning the diff) - Both fail open: if prior reviews cannot be fetched, behave as today - Config + action.yml inputs in the same PR; README, CLAUDE.md, Release-Notes.md; tryke tests (round counting, cap behaviour, low-only routing, fail-open) Refs the wave plan on #79.

agent-pi referenced this issue

2026-10-16 23:40:57 +00:00

Add verify and min_severity inputs to action.yml #108

agent-pi referenced this issue

2026-10-17 01:11:56 +00:00

Feed the previous review thread back into re-reviews #109

agent-pi referenced this issue

2026-10-17 01:59:00 +00:00

Skip unchanged files on re-reviews #110

agent-pi referenced this issue from a pull request that will close it,

2026-10-17 02:28:57 +00:00

Convergence controls: review-round cap and low-only wave soft stop #111