session-reflection
Session Reflection
Value: Feedback -- every user intervention is a signal that the system prompt is incomplete. Turning corrections into durable instructions creates compound improvement across sessions.
Purpose
Teaches agents to analyze session history for recurring corrections, generate project-specific instructions that prevent known failure modes, and maintain working state that survives context compaction and crashes. Transforms reactive corrections into proactive prevention.
Practices
Reflection Triggers
Reflect after: milestones (PR merged, feature complete), 3+ repeated corrections from the user, session restart or crash recovery, every 5 completed tasks, and on explicit user request. Do not wait for a "good time" -- reflect when triggered.
Analyze Session History
Examine conversation history, git log, memory files, WORKING_STATE.md, and session logs. Categorize each user intervention into one of five types:
- Correction: Agent did the wrong thing (instruction gap)
- Repetition: Agent was told the same thing again (emphasis gap)
- Role Redirect: Agent stepped outside its role (boundary gap)
- Frustration Escalation: User became more forceful (decay problem)
- Workaround: User did it themselves (skill gap)
See references/analysis-framework.md for detailed categorization and
prioritization.
Generate or Refine Project Instructions
Route project-specific directives into the appropriate instruction file based on content type. There is no separate system prompt file — everything flows from CLAUDE.md → AGENTS.md → .team/coordinator-instructions.md:
- CLAUDE.md — Session management (startup procedure, compaction recovery, state tracking), harness-specific configuration
- AGENTS.md — General project rules, dos/don'ts, coding conventions, workflow configuration
- .team/coordinator-instructions.md — Coordinator/pipeline-controller role distinctions, build pre-flight gates, spawn discipline, domain review checklists
Refinement rules: add new items for new gaps, promote advisory to structural
when gaps recur, rewrite ambiguous items for clarity. Never remove items
until the gap is confirmed solved across 3+ sessions. See
references/system-prompt-patterns.md for writing patterns and
references/launcher-templates.md for harness-specific file routing.
Self-Reminder Protocol
See CONSTRAINT-RESOLUTION.md in the template directory for the
consolidated self-reminder protocol (frequency, combined re-read list,
and post-compaction rules).
Working State Persistence
Maintain WORKING_STATE.md as insurance against context compaction and crashes.
Update after every significant state change. Always read before acting after
any interruption. Location: .factory/WORKING_STATE.md (pipeline) or project
root (standalone). See references/working-state-schema.md for format.
Do:
- Update on task start, phase change, decision made, blocker encountered
- Overwrite with current state (not append)
- Read FIRST after any interruption
Do not:
- Guess state from memory after a compaction or restart
- Use as a journal -- keep it concise and current
- Skip updates because "nothing important changed"
Post-Session Learning Loop
At session end: identify patterns from this session, update memory files, refine project instructions if triggers were hit, archive working state.
The loop closes when the same category of intervention stops recurring. If an intervention category persists across 3+ sessions after instruction refinement, escalate: the gap may require a new skill or a structural change to the workflow.
Enforcement Note
Advisory in all modes. Reflection triggers and instruction generation are self-enforced.
Hard constraints:
- Mandatory after context compaction:
[H]-- context compaction destroys implicit state; re-reading is not optional.
Constraints
- "3+ repeated corrections": A "repeated correction" means the user corrected the same underlying behavior multiple times. Different phrasings of the same correction count as one correction repeated. Different corrections in the same category (e.g., two different formatting preferences) count separately.
- "Never remove instructions until confirmed solved across 3+ sessions": This prevents premature cleanup. An instruction that hasn't triggered in 3 sessions might be preventing the problem -- that's success, not staleness. Remove only when you have positive evidence the underlying behavior is fixed (e.g., the code pattern that caused the issue no longer exists).
Verification
After completing work guided by this skill, verify:
- Reflection performed at every trigger point (milestone, repeated correction, restart)
- User interventions categorized using the five-type taxonomy
- Project instructions refined with structural (not just advisory) language
- Directives routed to correct file (CLAUDE.md / AGENTS.md / coordinator-instructions.md)
- Self-reminder protocol followed (state re-read every 5-10 messages)
- WORKING_STATE.md current and accurate
- State re-read after every context compaction (not guessed)
If any criterion is not met, revisit the relevant practice before proceeding.
Dependencies
This skill works standalone. For enhanced workflows, it integrates with:
- memory-protocol: Persistent storage for session learnings and working state
- agent-coordination: Coordination patterns referenced in instruction generation
- pipeline: Pipeline controller benefits from self-reminder and crash recovery
- ensemble-team: Team retrospectives feed into session reflection analysis
Missing a dependency? Install with:
npx skills add jwilger/agent-skills --skill memory-protocol