Session Protocol

"AI sessions start with zero context. Without a protocol, the first 10 minutes are wasted re-establishing what was done."

TL;DR

Implements the mandatory session initialization sequence from the Constitution: (1) load context files in order, (2) recover state from recent activity, (3) propose closure for pending items, (4) suggest next steps. Ensures every session starts productive from minute one, with full awareness of project state, open tasks, and recent decisions. [EXPLICIT]

Procedure

Step 1: Discover — Context Loading

Load in order (read each file):

1. CLAUDE.md — project instructions and agent rules
2. CONSTITUTION.md or references/ontology/constitution-v5.2.0.md — governance
3. insights/README.md — insights index (load domain files on-demand)
4. changelog.md — recent changes and decisions
5. tasklog.md — open tasks and pending work

Step 2: Analyze — State Recovery

After loading context:

1. Read last 5 changelog.md entries — understand recent session outcomes
2. Read all open items in tasklog.md — identify pending work, blockers, stale tasks
3. Check insights/ for any insights tagged tentative or needs validation
4. Check git status and recent commits on current branch
5. Check .specify/ for any in-progress specs, plans, or debates

Step 3: Execute — Pending Closure

Before accepting new work:

1. List all open tasks from tasklog.md with their age
2. For each: recommend close (done), continue (active), defer (deprioritize), or archive (no longer relevant)
3. Flag stale items (>7 days without progress)
4. Present summary to user — confirm before closing or archiving

Step 4: Validate — Next Steps Proposal

After pending items resolved:

1. Analyze current feature state (IIKit dashboard, branch status, phase progress)
2. Suggest 2-3 concrete next steps ranked by impact
3. Include at least one improvement beyond current task (from insights gaps, constitution TODOs, or observed patterns)
4. Wait for user direction — never start work without explicit confirmation

Quality Criteria

• All 5 context files loaded in correct order
• Last 5 changelog entries reviewed
• All open tasklog items identified
• Stale items (>7 days) flagged
• Pending closure recommendations provided
• 2-3 next steps proposed with rationale
• No work started without user confirmation

Anti-Patterns

Anti-Pattern	Why It's Bad	Do This Instead
Skipping context loading	Repeat past mistakes, miss decisions	Always load all 5 files
Starting work before closure	Task accumulation, invisible debt	Close pending items first
Only proposing continuation	Misses improvement opportunities	Include 1 innovation step
Auto-closing without user consent	User loses control	Always confirm closures

Related Skills

• continuous-learning — Insights loaded during context recovery
• tasklog-management — Tasklog is primary input for pending closure
• changelog-management — Changelog is primary input for state recovery

Usage

Example invocations:

• "/session-protocol" — Run the full session protocol workflow
• "session protocol on this project" — Apply to current context

Assumptions & Limits

• Assumes access to project artifacts (code, docs, configs) [EXPLICIT]
• Requires English-language output unless otherwise specified [EXPLICIT]
• Does not replace domain expert judgment for final decisions [EXPLICIT]

Edge Cases

Scenario	Handling
Empty or minimal input	Request clarification before proceeding
Conflicting requirements	Flag conflicts explicitly, propose resolution
Out-of-scope request	Redirect to appropriate skill or escalate