fix-findings

name	fix-findings
description	Acts on review panel findings by dispatching concurrent GPT 5.5 workers with medium reasoning effort for small-blast-radius items (one per file), then a GPT 5.5 worker with high reasoning effort for remaining issues with knock-on effects. All agents read relevant context files. Use after /review-panel produces findings.
allowed-tools	Read, Glob, Grep, Bash, Agent
argument-hint

Fix Findings

Triage review panel findings and dispatch agents to fix them. Coordinate — don't produce.

Agent brief (provide to every agent)

• The specific findings to address (file:line, problem, fix)
• Read context/lib/index.md and any files the router points to for the relevant area
• Read context/lib/context_style_guide.md before updating any comments or docs
• Read context/lib/development_guide.md before writing code
• Run cargo build and cargo test before considering the task done

Process

1. Triage

Classify each finding from the review panel output:

Small blast radius — GPT 5.5 medium reasoning effort, concurrent:

• Confined to a single file
• No interface or contract changes
• No knock-on effects in other packages
• Examples: missing error handling, nit, stale comment, dead code

Everything else — GPT 5.5 high reasoning effort, sequential:

• Crosses file or package boundaries
• Interface, contract, or exported type changes
• Knock-on effects likely
• Requires architectural judgment

Group small findings by file. Each file gets one agent.

2. Medium-reasoning workers (parallel)

Spawn one worker agent per file in a single message. Use model: "gpt-5.5" with reasoning_effort: "medium". Provide the agent brief above.

3. Wait and assess

Review outputs. Note unresolved findings.

4. High-reasoning workers (sequential)

Spawn 1–2 worker agents, one at a time. Use model: "gpt-5.5" with reasoning_effort: "high". Provide the agent brief, plus an enumeration of likely knock-on targets.

5. Report

What was fixed, what was skipped and why, and whether cargo build and cargo test pass.