Review Work

Strategic review using subagent for analysis, preserving main context for action decisions.

Git root: !git rev-parse --show-toplevel

Step 1: Determine Scope

Based on $ARGUMENTS:

• No args: Review current branch vs base
• --session: Review work captured in current session context
• Topic/area: Focus review on specific aspect

Step 2: Gather Context

• Read session context from <git-root>/.bonfire/index.md
• Get branch diff: git diff main...HEAD (or appropriate base)
• Read relevant specs/docs from .bonfire/
• Understand intent: what were we trying to accomplish?

Step 3: Run Review (Subagent)

Progress: Tell the user “Reviewing work for blindspots and gaps…”

Use the Task tool to invoke the work-reviewer agent.

Provide the review context:

Review this work for blindspots, gaps, and improvements.

**Scope**: [Branch diff / session work / specific area]

**Intent**: [What we were trying to accomplish - from session context]

**Files changed**:
[List of modified files from git diff]

**Session context**:
[Relevant notes from index.md]

Return categorized findings with severity and effort estimates.

Wait for the subagent to return findings before proceeding.

The subagent runs in isolated context (sonnet model), preserving main context for action decisions.

Review Validation

After the subagent returns, validate the response:

Valid response contains:

• ## Summary with finding counts
• At least one of: ## Fix Now, ## Needs Spec, or ## Create Issues

On valid response: Proceed to Step 4.

On partial response (missing categories but has findings):

1. Warn user: “Review returned partial results. Some categories may be missing.”
2. Present available findings.
3. Continue to Step 4.

On invalid/empty response:

1. Warn user: “Review subagent returned no findings. I’ll review directly.”
2. Fall back to in-context review:
   • Read changed files from git diff
   • Analyze for common issues (missing tests, error handling, etc.)
   • Check for TODOs and incomplete implementations
3. Present in-context findings using the same format.

On subagent failure (timeout, error):

1. Warn user: “Review subagent failed. Continuing with direct review.”
2. Perform in-context review as above.

Resumable Review (Large Changes)

For large changesets, review may need multiple passes. The Task tool returns an agentId you can use to resume.

When to offer resume:

• Review covers many files and user wants deeper analysis of specific area
• Findings mention “additional items omitted” or suggest areas need more attention
• User wants to focus review on specific aspect (e.g., “look more at tests”)

To resume review:

1. Tell user: “Review covered [X]. Want deeper analysis of [specific area]?”
2. If yes, re-invoke work-reviewer with the resume parameter:
   • Pass the agentId from the previous invocation
   • Provide a focused directive: “Continue review focusing on: [specific area].”
3. Merge findings from resumed review with previous findings.
4. Repeat if needed, up to 3 passes maximum.

Step 4: Present Findings

Present the subagent’s findings grouped by recommended action:

Fix Now (trivial effort)

[List items from subagent that can be fixed immediately]

Ask: “Want me to fix these now?”

Needs Spec (important, needs planning)

[List items that need implementation planning]

Ask: “Want me to create an implementation spec?”

Create Issues (large effort or nice-to-have)

[List items for future sessions]

Ask: “Want me to create GitHub/Linear issues?”

Step 5: Execute Chosen Action

Based on user choice:

• Fix now: Make the changes directly
• Spec: Run /bonfire-spec with findings
• Create issues: See below

Creating Issues

First, read <git-root>/.bonfire/config.json and check linearEnabled.

Offer choices based on config:

• Always offer: “Create GitHub issue”
• If linearEnabled is true: Also offer “Create Linear issue”

GitHub Issue Creation:

gh issue create --title "Finding title" --body "Finding details"

Linear Issue Creation (if enabled):

1. Use linear-cli to create the issue:

   linear issue create --title "Finding title" --description "Finding details"
   

2. On success: Return issue URL/ID
3. On failure: Warn user, offer to create GitHub issue instead

For each created issue:

• Record the issue ID and URL
• Note which tracker (GitHub/Linear) was used

Step 6: Update Session Context

Add review outcomes to <git-root>/.bonfire/index.md:

• Key findings noted
• Issues created (with links)
• Work deferred to future sessions