Pwrl End Session

PWRL End Session

Preserve session context with a well-documented commit capturing state, decisions, and next steps.

Interaction Method

• Use platform's ask_user_question, ask_user, ask_user_input, vscode/askQuestions or any available extension/tool for user interaction for all decisions
• Ask one question at a time
• Use multiple-choice questions when possible
• If input is empty, ask: "Would you like to end this session? Optionally provide a reason or context for why this session is ending."
• Provide clear recovery suggestions when errors occur

Purpose

End a work session cleanly by creating a single, clear commit that captures progress and context for the next session:

• Verify repository state and summarize changes
• Draft a descriptive commit message with decisions and next steps
• Create commit with mandatory [AGENT: ...] attribution
• Optionally extract learnings from the session work

Usage

/pwrl-end-session                           # End current session
/pwrl-end-session "switching to bugfix"     # End with reason note

After commit succeeds, optionally chain to /pwrl-learnings to capture session insights.

Key References

• checkpoint-protocol.md — Detailed validation rules for Phase 1
• commit-protocol.md — Message drafting rules and version bump handling
• commit-message-examples.md — Common session completion types
• pwrl-learnings — Optional chaining for extracting session insights

Architecture

Pure Skill Pipeline — Direct sequence of 2 micro-skills (checkpoint and commit), optionally followed by learnings extraction. The interactionMode is set in Phase 1 (checkpoint) and propagated through the artifact to Phase 2 (commit) and the optional Phase 3 (learnings chain):

INPUT (user invokes /pwrl-end-session)
  ↓
PHASE 1: Checkpoint (pwrl-end-session-checkpoint)
  → Step 1.5: Select Interaction Mode (detailed | smart | yolo)
  → Verify state, confirm completion
  → Output: Checkpoint artifact (includes interactionMode)
  ↓
PHASE 2: Commit (pwrl-end-session-commit)
  → Reads interactionMode from checkpoint artifact
  → Adjusts commit-message draft + approval flow
  → Output: Commit artifact
  ↓
PHASE 3 (Optional): Chain to pwrl-learnings
  → /pwrl-learnings-extract re-asks the mode for the learnings workflow
  → Extract and save session insights
  ↓
OUTPUT (session complete)

Interaction Mode Propagation

The interactionMode is set in pwrl-end-session-checkpoint Step 1.5 and controls behavior across the rest of the pipeline:

• detailed — User sees the draft commit message in pwrl-end-session-commit and edits it before approval; pre-flight summary in the checkpoint is shown line-by-line. Maximum control.
• smart — User sees a pre-flight summary (files, line counts, version-bump check) and approves the commit with one click; only pause for HIGH-risk operations (e.g., version bump detected, breaking-change warning). v1 simplification: behaves like Yolo with a single confirmation prompt at workflow start.
• yolo — Entire session-end (checkpoint + commit) auto-runs and only reports the final commit SHA. Fastest.

The mode set in this workflow does not affect Phase 3 (learnings chain). The optional chain to /pwrl-learnings re-asks the mode via pwrl-learnings-extract Step 1.5 so users can mix modes across phases (e.g., Yolo for the commit, Detailed for the learnings review).

Workflow: 3-Phase Pipeline

Each phase executes sequentially. The orchestrator invokes the micro-skill, validates output with quality gates, and passes the artifact to the next phase.

Phase 1: Checkpoint (pwrl-end-session-checkpoint)

Verify repository state and get session completion confirmation.

See detailed workflow: checkpoint-protocol.md

• Check working tree state (git status)
• Identify all changes (staged, unstaged, untracked)
• Display summary and get user confirmation
• Handle incomplete work (capture next steps)
• Generate checkpoint artifact

Phase 2: Create Commit (pwrl-end-session-commit)

Prepare commit message and create commit with proper attribution.

See detailed workflow: commit-protocol.md

• Draft subject (imperative, ≤50 chars)
• Write body (why/what/next)
• Detect version bump and update CHANGELOG.md if needed
• Get user approval of message
• Stage files and create commit
• Capture commit SHA

Phase 3: Document Learnings (Optional Chain)

Capture session insights and learnings for future reference.

Invoke: /pwrl-learnings with changed files from Phase 2

• User can opt in or skip
• Extracts, classifies, deduplicates, and saves learnings
• Links learnings to session commit SHA
• Result: Learning documents in docs/learnings/

Rules

• ✓ Verify working tree before starting
• ✓ Agent trailer mandatory: [AGENT: ...] on last line
• ✓ No automatic push (user controls push timing)
• ✓ User approval required for commit message
• ✓ Changelog updated on version bump
• ✓ Learnings extraction is optional, user decides

Best Practices

• Make incomplete work actionable: list specific next steps in body
• Link to task/plan file if applicable (e.g., "Completed: docs/tasks/...")
• Keep message readable (wrap at ~72 chars)
• Include context about why work ended here (partial, blocked, switching focus)

Acceptance Criteria

• Input: User confirms session completion and there are changes
• Output: Created commit with [AGENT: ...] trailer and descriptive body
• Version bump: Updated CHANGELOG.md staged in commit if version changed
• Verification: Commit SHA displayed; learnings extraction offered as optional next step