Repository navigation
review: read and output budget on the small-diff fast path - #1964
Closed
remigiusz-antczak wants to merge 1 commit into
Closed
remigiusz-antczak wants to merge 1 commit into
remigiusz-antczak wants to merge 1 commit into
Conversation
One-line BUDGET rule in deep-code-review SKILL.md: git diff --stat -p as the one patch read, grep callers and callees then offset/limit reads, tail -n 40 on test and CI output, no narration. Security checks unchanged. Pinned by scripts/test_review_read_budget.py and a new eval case. Tests: 8/8 (test_review_read_budget 1, test_always_loaded_bytes 6, mustload gate, size-ratchet gate) Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_014QNFFPWVcPUbeuvwM5fk2Z
Collaborator
Author
|
Superseded by #1982. |
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Adds a read and output budget to deep-code-review's small-diff fast path, aimed at the review-cost gap (the review bench ties its rival at about 3.5x the cost).
Measured local split of review-session size: Read results ~47%, Bash output ~27%, model output ~22%. Skill bytes are under 1% (see #1963). This targets the first two.
SKILL.md.git diff --stat -pis the single patch read. The reviewer greps the callers and callees of each changed symbol and reads those ranges plus hunk context with offset/limit. A whole file is read only if it is under ~200 lines. Test and CI output is piped throughtail -n 40, with no narration between tool calls. The line states it never skips a security check.scripts/test_review_read_budget.pypins the rule's parts and its place in the fast path, and checks that steps 1 and 5 (the security and guarantee floor) still exist. A new eval case is added inevals/evals.json.Not yet measured: the cost and recall effect on the review bench. The expected Read and Bash cut is an estimate. A paired bench run is the next step before this counts as a cost win.
Tests: 8/8 (read-budget 1, always-loaded bytes 6, mustload gate). Independent review: 0🔴; the 3🟡 and 1🔵 are folded in. The main one is that the first draft did not let the reviewer read callers.
🤖 Generated with Claude Code
https://claude.ai/code/session_014QNFFPWVcPUbeuvwM5fk2Z