Skip to content

fix(#462): add no-content-abbreviation directive to pr-review skill - #463

Closed
fullsend-ai-coder[bot] wants to merge 1 commit into
mainfrom
agent/462-no-content-abbreviation
Closed

fullsend-ai-coder[bot] wants to merge 1 commit into
mainfrom
agent/462-no-content-abbreviation

Conversation

@fullsend-ai-coder

Copy link
Copy Markdown
Contributor

Summary

  • Adds a "No content abbreviation" directive to step 3d (context package assembly) of the pr-review skill (skills/pr-review/SKILL.md)
  • Explicitly prohibits the orchestrator from abbreviating, summarizing, or replacing source file content with placeholder strings (e.g., '...', [truncated]) when composing sub-agent context packages
  • Provides guidance for handling files too large to include: use explicit boundary markers for partial inclusion, or omit entirely and list as unavailable

Motivation

During Review Run 2 on PR #326, the orchestrator abbreviated test fixture JSON content with placeholder strings. The correctness sub-agent misinterpreted these placeholders as actual test data and raised a false positive [test-integrity] finding. The challenger caught it, but the pattern is a systematic risk — sub-agents cannot distinguish abbreviated content from genuine file contents.

Test plan

  • Verify the directive appears in step 3d of skills/pr-review/SKILL.md between the context package field list and section 3e
  • On the next 10 review agent runs on PRs with test fixtures, monitor for zero instances of false positives caused by abbreviated or placeholder content in context packages

Closes #462

Post-script verification

  • Branch is not main/master (agent/462-no-content-abbreviation)
  • Secret scan passed (gitleaks — 3760b3bb70de32596c1922332fb96e76cd9ebcd8..HEAD)
  • PR body secret scan passed (gitleaks — no-git)
  • Pre-commit hooks passed (authoritative run on runner)
  • Tests ran inside sandbox

@fullsend-ai-coder
fullsend-ai-coder Bot requested a review from a team July 27, 2026 08:28
Add an explicit instruction in step 3d (context package assembly)
of the pr-review skill prohibiting the orchestrator from
abbreviating, summarizing, or replacing source file content with
placeholder strings when composing sub-agent context packages.

The review orchestrator was observed abbreviating test fixture
JSON content with placeholders (e.g., '...'), causing the
correctness sub-agent to misinterpret the placeholders as actual
test data and raise false positive [test-integrity] findings.
While the challenger caught this instance, the pattern is a
systematic risk. This directive serves as a guardrail against
the model's tendency to compress content during prompt
composition.

Closes #462
@ben-alkov
ben-alkov force-pushed the agent/462-no-content-abbreviation branch from 99cea7c to 98ee7ce Compare July 30, 2026 23:25
@ben-alkov ben-alkov added review-agent ready-for-review Triggers review agent dispatch bug Something isn't working labels Jul 30, 2026
@fullsend-ai-review

fullsend-ai-review Bot commented Jul 30, 2026 •

Copy link
Copy Markdown

🤖 Finished Review · ✅ Success · Started 11:28 PM UTC · Completed 11:36 PM UTC
Commit: 98ee7ce · View workflow run →

@fullsend-ai-review

Copy link
Copy Markdown

Review

Findings

Medium

@fullsend-ai-review fullsend-ai-review Bot added the requires-manual-review Review requires human judgment label Jul 30, 2026
@github-actions

Copy link
Copy Markdown

This pull request has been automatically marked as stale because it has not had any activity in the last month. It will be closed in 2 weeks if no further activity occurs. Remove the stale label to reset the inactivity timer.

@github-actions github-actions Bot added the stale label Aug 30, 2026
@github-actions

Copy link
Copy Markdown

This pull request has been closed due to inactivity. Feel free to reopen it if the changes are still needed.

@github-actions github-actions Bot closed this Sep 14, 2026
@github-actions
github-actions Bot deleted the agent/462-no-content-abbreviation branch September 14, 2026 01:49
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

bug Something isn't working ready-for-review Triggers review agent dispatch requires-manual-review Review requires human judgment review-agent stale

Projects

None yet

Development

Successfully merging this pull request may close these issues.

Review orchestrator should not abbreviate source file content with placeholder strings in sub-agent context packages

1 participant