Resolve current state: explicit active task, session → docs, clear/archive on finish

[jdugarte]

Summary

This PR fixes cross-IDE and cross-session “wrong stage” behavior by making the active task explicit in agentic:guild memory, absorbing session knowledge into project docs (sync-docs and harvest-rules), and optionally clearing/archiving the task when the branch is finished (with user confirmation). It also applies Bugbot-driven fixes (hard constraints, yields, directory list, and status-check/sync-docs/harvest-rules edge cases).

Changes

1. Explicit active task (status-check, start-task, explore-task, finish-branch, rules)

• skills/status-check/SKILL.md: Discovers sessions, resolves active task from <active_task_pointer>. If pointer is missing/ambiguous with multiple sessions, lists sessions and asks user to confirm or set the pointer. Handles stale pointer (file missing) and zero sessions + [NONE].
• skills/start-task/SKILL.md, skills/explore-task/SKILL.md: Set <active_task_pointer> when creating a session so the active task is unambiguous across IDEs.
• skills/finish-branch/SKILL.md: Retains pointer in step 3.1; in steps 4.4–4.5, after branch is finished, asks user to clear the active task and archives the session file to .agenticguild/completed_sessions/ (with date suffix), then sets pointer to [NONE].
• templates/core/AGENTIC_GUILD_RULES.md: State-tracking rule requires writing the active session filename to <active_task_pointer> when a task is active.
• templates/core/memory_scaffold/current_state.md: Comment added that when multiple session files exist, set the pointer for cross-IDE consistency.
• playbooks/EXPECTED_PROJECT_STRUCTURE.md: current_state and completed_sessions documented; sync.sh and skills/update-agentic-guild/SKILL.md create/ensure completed_sessions and its .gitkeep.

2. Session memory → docs (sync-docs, harvest-rules)

• skills/sync-docs/SKILL.md: Uses two inputs — branch diff and active task session (when present). Phase 1.1 (Variant B): reads session, semantically analyzes full content, classifies by knowledge type, merges with diff-based “needs update” list; report includes source (diff | session | both). Phase 3.1: applies both; SCHEMA_REFERENCE is enriched with session-derived “why/when” of data; conventions/patterns go to .cursorrules and docs/core/SYSTEM_ARCHITECTURE.md. Knowledge-type map and session_to_docs block added.
• skills/harvest-rules/SKILL.md: Step 1.0 reads active session and extracts rule-worthy content; step 1.1 merges session-derived candidates with diff and review_ledger. Dedup: skip candidates that duplicate existing rules or that are semantically equivalent to what sync-docs likely just wrote when run after sync-docs in the same workflow.
• playbooks/EXPECTED_PROJECT_STRUCTURE.md: Docs to Sync table includes .cursorrules (conventions/patterns from diff or session). Note added that both sync-docs and harvest-rules write to .cursorrules by design and how to avoid near-duplicates.
• playbooks/AI_WORKFLOW_PLAYBOOK.md: Finish-branch and skills table updated to describe session-based doc sync and harvest.

3. Bugbot / consistency

• Cherry-picked fd18210 (previous lingering bugbot): <hard_constraints> (no git execution) in audit-compliance, harvest-rules, pr-description, process-feedback, roadmap-consult, roadmap-manage, status-check; explicit <yield> in harvest-rules 1.0 and process-feedback 1.1–1.3; process-feedback in update-agentic-guild skill list; update-agentic-guild route in templates/core/AGENTIC_GUILD_RULES.md.
• Status-check: handling for stale pointer and for zero sessions with [NONE] pointer.
• update-agentic-guild: completed_sessions added to step 1.1 directory list.

Files touched

playbooks/AI_WORKFLOW_PLAYBOOK.md, playbooks/EXPECTED_PROJECT_STRUCTURE.md, skills/explore-task/SKILL.md, skills/finish-branch/SKILL.md, skills/harvest-rules/SKILL.md, skills/start-task/SKILL.md, skills/status-check/SKILL.md, skills/sync-docs/SKILL.md, skills/update-agentic-guild/SKILL.md, sync.sh, templates/core/AGENTIC_GUILD_RULES.md, templates/core/memory_scaffold/current_state.md.

[cursor]

PR Summary

Medium Risk
Updates core agentic:guild workflow state handling and doc-sync/rule-harvest behavior; mistakes could cause the assistant to pick the wrong active task or write updates to the wrong docs. No application/runtime code is changed, so impact is limited to developer tooling/automation.

Overview
Makes .agenticguild/current_state.md authoritative by requiring <active_task_pointer> to be set whenever a task is active, and updates start-task, explore-task, finish-branch, and status-check to preserve/resolve that pointer (including safer handling when multiple sessions exist).

Extends sync-docs and harvest-rules to incorporate the active task session (plus review_ledger) as an input: session content is semantically classified and used to propose/update SPEC, ADRs, SCHEMA_REFERENCE enrichment, and .cursorrules, with guidance to avoid near-duplicate rule suggestions.

Adds an explicit finish-branch flow to optionally clear the active task and archive the session to .agenticguild/completed_sessions/, and updates bootstrap/update tooling (sync.sh, update-agentic-guild, project structure docs, and templates) to create/track this directory and to forbid skills from running git commit/push/merge via tools.

^{Written by Cursor Bugbot for commit 2c43269. This will update automatically on new commits. Configure here.}

[cursor]

Cursor Bugbot has reviewed your changes and found 2 potential issues.

^{Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}