devloop-analyze

name	devloop-analyze
description	Use when running devloop council analysis on a repo. Covers the correct op run invocation, model selection, and narrative output format. Invoke instead of `just analyze` which is broken.

devloop-analyze

The Problem with just analyze

The just analyze recipe is broken because it doesn't handle env conflicts from shell-injected op:// vars.

Correct Invocation

# The op-conflict-resolver hook will auto-detect conflicts and suggest the prefix.
# Manual form if needed:
env -u TAVILY_API_KEY -u CONTEXT7_API_KEY -u GEMINI_API_KEY -u FIRECRAWL_API_KEY \
  op run --account=my.1password.com --env-file=/Users/joe/.secrets -- \
  devloop analyze --repo /path/to/repo

Note: If op-conflict-resolver hook is active, just run op run --account=my.1password.com --env-file=/Users/joe/.secrets -- devloop analyze --repo /path/to/repo and the hook will correct it automatically.

Model Selection

• Use: OpenAI models (gpt-4o, gpt-4o-mini)
• Avoid: Local qwen3 — returns 404, unreliable
• If devloop has a --model flag, specify explicitly

Output Format Preference

When presenting council results:

• Format: Narrative timeline — one prose paragraph per time block (e.g., "Morning session", "Afternoon")
• NOT: Bullet point lists of commits
• Tone: Conversational, compact — describe what happened and why, not just what changed
• Example: "Tuesday morning was focused on stabilizing the BAML codegen pipeline — three commits in quick succession fixed the version mismatch between Cargo.toml and generators.baml that had been blocking the type-safe client generation."

Pieces LTM Context

Before or after devloop analyze, query Pieces for historical context:

ask_pieces_ltm: "what was I working on in [repo name] recently?"
ask_pieces_ltm: "what problems did I hit with [specific feature]?"

Requires PiecesOS running: curl -s http://localhost:39300/.well-known/health

Devloop Export

After analyze, export for sharing:

devloop export --repo /path/to/repo --output /tmp/report.md

Common Failures

Symptom	Fix
op run resolves wrong secret	Shell has conflicting var — use env -u or let hook auto-fix
Council returns no results	Check --repo path is correct git repo
qwen3 model errors	Switch to OpenAI model explicitly
PiecesOS not responding	curl http://localhost:39300/.well-known/health — restart if down