Repository navigation
feat: add a sonnet task-reviewer for plan-loop reviews and drop "when in doubt, opus" - #58
Merged
Merged
Conversation
… in doubt, opus" Superpowers subagent-driven-development scales the model of its per-task reviews and scoped re-reviews to the diff, with a mid-tier floor, while the anchor sent every review to the opus-pinned reviewer. Sessions resolved the conflict by dispatching reviewer with model=sonnet, which the stats report flags as below pin: 69 of about 230 reviewer dispatches in 11 days on one machine, 6 of 38 on another. task-reviewer (sonnet/high) takes those seats and flags diffs that need reviewer. reviewer stays opus for final, whole-branch, high-risk and pre-PR review, so a bare reviewer dispatch still fails toward the stronger model. The implementer line loses "when in doubt, opus". In 199 implementer dispatches over the same window, 104 ran on opus and about 98 of those named no escalation reason; sonnet tasks showed no sign of extra fix rounds (0.16 fix dispatches per task against 0.21 on opus, confounded by task size). The explicit escalation categories stay.
AqueGen
added a commit
that referenced
this pull request
Oct 13, 2026
…d give haiku agents real effort (#57) * fix: price Haiku 5.5 per request, Sonnet 5.5 cache reads at 0.05x, and give haiku agents real effort Haiku 5.5 is what `haiku` resolves to on the Anthropic API since Claude Code 2.1.293. The stats report had no price row for it, so its volume showed as unpriced. Haiku 5.5 bills each request on one of two cards by prompt length (cache reads and writes included): $0.10/$0.50 up to 100K tokens, $0.50/$2.50 above. The card is now decided per deduped usage line and kept as a long-prompt share next to the totals, so costOf prices each share on its own card. Every flat-rate model prices exactly as before. Sonnet 5.5 reads cache at 0.05x ($0.10/MTok), not 0.1x, so opus 5.5 to sonnet 5.5 is a flat 50% on every token type and the cache-parity argument in SKILL.md and README is gone. Haiku 5.5 supports the full effort ladder with a Claude Code default of medium, so the haiku agents' effort pins now take effect on the API. surveyor and verifier move from low to medium, since at low Haiku 5.5 "is more likely to skip a search, stop early, or skip a check". test-runner stays low. On the cloud platforms `haiku` is still Haiku 4.5 with no effort knob; the docs now say so. * feat: add a sonnet task-reviewer for plan-loop reviews and drop "when in doubt, opus" (#58) Superpowers subagent-driven-development scales the model of its per-task reviews and scoped re-reviews to the diff, with a mid-tier floor, while the anchor sent every review to the opus-pinned reviewer. Sessions resolved the conflict by dispatching reviewer with model=sonnet, which the stats report flags as below pin: 69 of about 230 reviewer dispatches in 11 days on one machine, 6 of 38 on another. task-reviewer (sonnet/high) takes those seats and flags diffs that need reviewer. reviewer stays opus for final, whole-branch, high-risk and pre-PR review, so a bare reviewer dispatch still fails toward the stronger model. The implementer line loses "when in doubt, opus". In 199 implementer dispatches over the same window, 104 ran on opus and about 98 of those named no escalation reason; sonnet tasks showed no sign of extra fix rounds (0.16 fix dispatches per task against 0.21 on opus, confounded by task size). The explicit escalation categories stay.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Stacked on #57. GitHub retargets it to
mainonce #57 merges.What changes
Superpowers subagent-driven-development picks the model for its per-task reviews and scoped re-reviews by diff size, with "a mid-tier model as the floor for reviewers". Our anchor sent every review to
reviewer, which is pinned to opus. Sessions resolved that by dispatchingreviewerwithmodel=sonnet, and the stats report flags those as below pin: 69 of about 230 reviewer dispatches in 11 days on my machine, and 6 of 38 on another user's. The transcript descriptions are nearly all "Review Task N (spec + quality)" and "Re-review Task N fix round 1".agents/task-reviewer.md(sonnet/high) for per-task reviews and scoped re-reviews in a plan loop, and for reviews explicitly asked for at the sonnet tier. It follows the caller's review template and says in its first line when a diff touches security, money, migrations, concurrency or public contracts, or is too large for one pass, so the caller can send it toreviewer.reviewerstays opus for final, whole-branch, high-risk and pre-PR review. We kept the bare default on opus on purpose: one "Final whole-branch review" in the transcripts went out with nomodel=, and re-pinningreviewerto sonnet would have run it on sonnet without anyone noticing.belowPinneeds no change.Testing
node --test hooks/dispatch-counter.test.mjs hooks/update-check.test.mjs: 127 pass, 0 fail, including the AGENT_PINS/frontmatter sync test.claude plugin validate .passed.