Session Retro

Session Retro

A session retrospective that turns conversation mistakes into durable process improvements, and successful patterns into reusable skills.

Why this matters

Without retro, the same mistakes repeat across sessions — Claude gets corrected, apologizes, and forgets by next conversation. Retro breaks this cycle by tracing errors to their source (a skill, CLAUDE.md, memory, or tool behavior) and proposing concrete fixes that persist.

Process

1. Scan conversation for correction signals
2. Classify: errors found → Path A | smooth session → Path B
3. For each finding: analyze → attribute → propose action
4. Run Path C (token efficiency analysis)
5. Present report with action proposals
6. Ask user which actions to execute
7. Ask user if they want to save the report

Step 1: Scan for Correction Signals

Search the conversation for moments where the user corrected or redirected Claude's behavior.

Correction signal patterns (not exhaustive — use judgment):

• Direct corrections: "wrong", "no", "that's not right", "incorrect"
• Redirections: "shouldn't it be...", "why did you use X instead of Y", "you missed..."
• Repeated instructions: user restating something they already said
• Frustration cues: "wrong again", "still failing", "I already said..."
• Questions challenging behavior: "why", "which process did you follow", "the skill says..."
• Explicit review requests: "review the session log", "trace what happened..."

Also scan for silent corrections — places where the user quietly fixed something Claude should have done (e.g., user manually ran a command Claude skipped).

Step 2: Classify the Session

• Errors found → Path A (analyze each error)
• No errors found → Path B (extract successful patterns)
• Mixed → Do both Path A and Path B After completing Path A and/or Path B, proceed to Path C (step 4).

Path A: Error Analysis

For each correction found, build a root cause analysis:

2a. What happened?

State factually: what did Claude do, and what should it have done? No defensiveness, no hedging. Just the facts.

2b. Trace the source

Identify which document or mechanism should have guided the correct behavior:

Read the actual source file to verify — don't rely on memory of what it says.

2c. Attribute the cause

Determine which category the error falls into:

1. Source is unclear — The document exists but is ambiguous or poorly structured, making it easy to misinterpret. (Example: a default rule buried in an "override" subsection)

2. Source is missing — No document covers this scenario. Claude had to improvise and guessed wrong.

3. Execution drift — The document is clear, but Claude didn't follow it. This happens when instructions are long or when shortcuts seem reasonable in context.

4. One-off mistake — A simple slip (wrong file path, typo, etc.) with no systemic cause.

2d. Propose action

Based on the attribution:

For each action, specify:

• Which file to modify (exact path)
• What to change (before/after or description)
• Why this fix prevents recurrence

Path B: Success Pattern Extraction

When the session went smoothly, look for patterns worth preserving:

1. Multi-step workflows that succeeded — could they become a skill?
2. Novel tool combinations — a sequence of tools that solved a problem efficiently
3. Decisions that avoided problems — defensive choices that paid off
4. Reusable subagent prompts — agent dispatches that returned good results

For each pattern, assess:

• Frequency: Will this come up again? (one-off → skip; recurring → extract)
• Complexity: Is it complex enough that a skill adds value? (trivial → skip)
• Brittleness: Would the pattern break without the skill guiding it? (fragile → extract)

Only propose extraction for patterns scoring high on at least 2 of 3 criteria.

Path C: Token Efficiency Analysis

Runs after Path A/B. The goal is to find token waste within this session and suggest concrete improvements for future sessions.

Claude Code cannot access exact token counts, so use heuristic estimation based on observable signals: file sizes read, tool call count, subagent dispatches, and conversation round-trips. Focus on actionable patterns, not precise numbers.

4a. Scan for Cost Signals

Review the conversation for these patterns: