Migrate .claude/ prompts for Claude Opus 4.7 (Phase 1)#4636
Draft
Migrate .claude/ prompts for Claude Opus 4.7 (Phase 1)#4636
Conversation
Opus 4.7 follows instructions more literally than 4.6. Phrases like "minimize false positives", "filter aggressively", and "only flag high-severity" — flagged in Anthropic's 4.7 migration guide — cause measurable recall drops when the agent's job is review or analysis. Reframe the four highest-impact prompts so agents surface uncertain or scope-adjacent findings instead of silently dropping them: - security-reviewer: add CONCERN status tier between PASS and FAIL, plus an "Other Security Observations" escape hatch for security-adjacent issues outside S0/S1/S2 (secret leaks, CSRF, injection, SSRF, etc.) - codebase-analyzer: remove blanket "DO NOT identify bugs / performance / security concerns" suppression; add an Observations section for clearly broken or risky code spotted while tracing - context-analyzer: replace "Filter Aggressively / Filter Ruthlessly" with prioritization framing that preserves rejected options and superseded workarounds when they carry decision rationale - validate-plan: stop pre-filtering deviations; let the reader triage Scope is intentionally narrow — only the Category C recall suppressors from the 4.7 audit. The Category A anti-laziness scaffolding (CRITICAL / MUST / ALWAYS / NEVER emphasis) and the broader pre-Opus-4 redundancy review will be addressed in follow-up passes.
Phase 1 Tier 2 of the Opus 4.7 prompt-migration audit. Opus 4.7 follows emphasis too literally — uppercase imperatives (CRITICAL / YOU MUST / ALWAYS / NEVER) coerce overtriggering, and forced pre-action checklists cause over-reporting. Anthropic's 4.7 migration guide flags these as the highest-volume patterns to rewrite. Reframe 17 files across four clusters, preserving all technical content (VM deadlock explanations, Tailwind disabled:hover mechanics, Y.Doc transaction rules, the git attribution safety rail): - Core instructions (CLAUDE.md + 5 guidelines): drop⚠️ /uppercase wrappers on real hazards; collapse duplicate "Summary of Critical Rules" and pre-submission / pre-POM-edit checklists into prose - Agent definitions (5 files): drop "You Must Follow" section headers and stacked DO-NOT/ONLY scope-suppression blocks (top and tail) in codebase-locator and codebase-pattern-finder; remove fixed time budgets from idea-machine steps - Slash commands (4 files): drop CRITICAL/FRESH/EXTREMELY stacks and the duplicate REMEMBER block in research-codebase; make progress tracking advisory rather than mandatory; keep commit.md's NEVER attribution rail as a legitimate safety rail - E2E nested guides (2 files): drop CRITICAL/MUST read-first block and 6-item Research Checklist in page-objects; soften "Always wait for connection" Tier 2 scope is intentionally limited to Category A (anti-laziness scaffolding), B (forced progress/checklist scaffolding), and D (fixed time/effort assumptions). Tier 1 (Category C recall suppressors) was committed earlier on this branch. The broader pre-Opus-4 redundancy audit is Phase 2.
Codecov Report✅ All modified and coverable lines are covered by tests. Additional details and impacted files@@ Coverage Diff @@
## main #4636 +/- ##
==========================================
- Coverage 89.63% 89.61% -0.02%
==========================================
Files 444 444
Lines 21558 21558
==========================================
- Hits 19324 19320 -4
- Misses 2234 2238 +4 ☔ View full report in Codecov by Sentry. 🚀 New features to boost your workflow:
|
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.
Description
Migrates
.claude/prompts and guidelines for Claude Opus 4.7 per Anthropic's 4.6 → 4.7 prompt-migration guide.Opus 4.7 follows instructions more literally than 4.6. Two behaviors matter most for our agent prompts:
CRITICAL:/YOU MUST/ALWAYS/NEVER) and forced pre-action checklists cause overtriggering and over-reportingThis is Phase 1 of a two-phase audit. Phase 2 (pre-Opus-4 content redundancy review) will follow separately.
Commits
2e95506— Tier 1: relaxed recall suppressors insecurity-reviewer,codebase-analyzer,context-analyzer,validate-plan(+70/-59, 4 files). Added aCONCERNstatus tier and an "Other Security Observations" escape hatch tosecurity-reviewer.64334b1— Tier 2: softened anti-laziness scaffolding and collapsed forced checklists across 17 files (+113/-198).Technical content preserved
Emphasis softened, hazard explanations kept verbatim:
yex-guidelines.mddisabled:hover:mechanics inui-patterns.mdreact-collab-editor.mdtesting-essentials.mdNEVER add co-author information) incommit.mdScope
Out of scope for this PR (deferred to Phase 2):
.claude/guidelines/testing/*.md) — audited as clean, no changes neededCloses #__
Validation steps
security-reviewersurfacing ambiguous findings asCONCERNrather than collapsing toPASScodebase-analyzernoting observations at the end of analyses when relevantAdditional notes for the reviewer
codebase-locator.md/codebase-pattern-finder.mdthat were partially undoing the top-of-file softeningAI Usage
Please disclose whether you've used AI anywhere in this PR (it's cool, we just
want to know!):
You can read more details in our
Responsible AI Policy
Pre-submission checklist