Skip to content

Conversation

@dguido
Copy link
Member

@dguido dguido commented Feb 12, 2026

Summary

  • Adds function-analyzer agent (Read, Grep, Glob) for ultra-granular per-function deep analysis
  • Pure context building only — no vulnerability identification
  • Includes per-function microstructure checklist, cross-function flow rules, quality thresholds, and anti-hallucination rules
  • Updates SKILL.md Section 8 to reference the agent
  • Bumps plugin version to 1.1.0

New files

  • plugins/audit-context-building/agents/function-analyzer.md (133 lines)

Test plan

  • YAML frontmatter parses correctly
  • {baseDir}/skills/audit-context-building/resources/ paths resolve to existing files
  • SKILL.md changes don't break existing skill trigger behavior
  • marketplace.json version matches plugin.json (both 1.1.0)

🤖 Generated with Claude Code

dguido and others added 2 commits February 11, 2026 21:56
Introduces a formal agent definition for ultra-granular per-function
analysis. Updates SKILL.md Section 8 to reference agent and bumps
version to 1.1.0.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
@dguido dguido requested a review from omarinuwa as a code owner February 12, 2026 03:09
- Wrap long SKILL.md line (239 chars -> multi-line) for readability
- Trim verbose agent description frontmatter
- Add "When NOT to Use" section to agent file per project standards

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
@dguido
Copy link
Member Author

dguido commented Feb 12, 2026

Code Review Summary

Findings by severity

  • P2 (important): 1
  • P3 (nice to have): 3 (1 fixed, 2 dismissed)
  • P4 (informational): 4

Fixed (3 total)

  1. P2: SKILL.md line 266 was 239 characters -- Wrapped to multi-line for readability. Pre-existing lines in the file go up to ~161 chars; this was notably longer.

  2. P3: Agent description frontmatter was 271 characters -- Trimmed redundant tail clause ("that require detailed understanding before vulnerability hunting") while preserving all trigger keywords. Now 196 chars.

  3. P3: Missing "When NOT to Use" section in agent file -- Added per project quality standards. Lists four clear negative boundaries: vulnerability work, high-level overviews, trivial functions, and code modification tasks.

Dismissed (2 total)

  1. P3: Content overlap between agent file and SKILL.md -- Intentional and necessary. Per Anthropic docs, subagents receive only their own system prompt, not the parent skill content. The agent must duplicate the checklist and rules to function correctly. The agent already uses progressive disclosure by referencing resource files for detailed examples rather than inlining them.

  2. P3: Agent lacks Bash tool access -- The restricted tool set (Read, Grep, Glob) is a deliberate design choice enforcing the read-only "pure context building" constraint. Adding Bash would allow arbitrary command execution, contradicting the agent's security-focused design.

Informational (P4)

  • Version bump 1.0.0 -> 1.1.0 is appropriate for a backwards-compatible feature addition
  • plugin.json and marketplace.json versions match (1.1.0)
  • {baseDir} paths resolve correctly to existing resource files
  • tools: frontmatter key is the correct key for agent files per Anthropic docs

Quality pipeline

All CI checks pass: JSON validation, marketplace consistency, SKILL.md frontmatter, no hardcoded paths, no personal emails, pre-commit hooks (ruff, shellcheck, shfmt, trailing whitespace, end-of-files).

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant