[efraespada]

Bugbot could recommend replacing a valid external Action input with an undeclared key based on another version or model memory. Review and the real-agent benchmark now share a compact instruction that checks the exact Action revision, manifest/runtime getter, forwarded inputs, defaults and effective values before reporting a defect or suggesting a replacement. Unverifiable claims abstain; verified wrong-key and ignored-value failures remain reportable.

Closes #739.

Fifteen new delivery, verdict-isolation, scorer and native YAML/compiled-condition controls extend the corpus to 71 cases. The owning SDD, catalog, detection/quality documentation and runtime bundles are updated. The original long-ignore-list prompt-size limit remains enforced; the final instruction is 713 characters.

Validation:

• 772 suites / 14,132 tests pass, 35 existing skips; all global/module coverage budgets and setup acceptance; 267 extension tests with 100% in all four metrics.
• 91 directed tests, including the original prompt-size regression; types, lint, specifications/catalog, docs/workflows/agent docs, Graphify, build/package 87 files and local CLI/API/web smoke pass.
• Local patch audit: 2 executable modified production lines covered; zero misses or uncovered changed branches. Normal commit hooks pass build/full Jest/lint.

Review evidence: the immutable upstream manifest declares problem-matcher; its runtime reads that input and disables registration for false. This is read-only source corroboration, separate from the consumer's reported SDK/production runs.

The actual measurements retain all four instruction variants, fixed inputs, identities, predictions and scores. The final sample reports all three real defects and no negative-case findings, but its full default score fails because two anchors are off by one and a category differs from the reference. The full-text intermediate passed model scoring but exceeded the existing context-size test; a compact intermediate missed the ignored-value case. Inputs, expectations and scoring thresholds are unchanged. Reference predictions validate the deterministic scorer; these optional measurements are not an all-thresholds-passed or statistical improvement claim. Model guidance remains probabilistic.

[docs-page]

To preview the documentation for this pull request, visit the following URL:

docs.page/vypdev/copilot~758

Documentation is deployed and generated using docs.page

[vypbot]

Bugbot: review complete

Current status: No active findings on cafdeb5.

Pull request · Verified commit · Workflow run

[vypbot]

Copilot approval · Human review needed

Status: A protected workflow, policy, or trust-boundary file changed.
Completed: 1/1 observed checks succeeded; Bugbot complete.
Action required: Ask a human reviewer to decide.
Impact: No new approval is confirmed for this revision.
See checks

Technical evidence

Revision: cafdeb518482; base: e45011390516; reason: protected-path.

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 94.79%. Comparing base (5b3664f) to head (cafdeb5).
⚠️ Report is 39 commits behind head on develop.

Additional details and impacted files

@@             Coverage Diff             @@
##           develop     #758      +/-   ##
===========================================
+ Coverage    94.50%   94.79%   +0.28%     
===========================================
  Files         1035     1078      +43     
  Lines        33299    34859    +1560     
  Branches      9824    10354     +530     
===========================================
+ Hits         31468    33043    +1575     
+ Misses         699      684      -15     
  Partials      1132     1132              

Files with missing lines	Coverage Δ
src/prompts/bugbot.ts	100.00% <100.00%> (ø)
src/tooling/bugbot_benchmark_runner.ts	77.77% <100.00%> (ø)

... and 43 files with indirect coverage changes

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.