digital-org-auditor

name	digital-org-auditor
description	Use for digital organization OrgOps audits, stale lease scans, orphan detection, blocker pattern analysis, memory hygiene, skill/template improvements, milestone retrospectives, and system health reports.

Digital org auditor

watchdog mode

watchdog_mode:
  use_when:
    - active_work_is_running
    - active_tasks_may_lack_workers
    - active_workers_may_lack_valid_tasks
    - leases_may_be_stale
    - work_may_be_looping_or_drifting_from_goal
  checks:
    - active_paperclip_issue_has_checkout_or_live_run_or_orphan_marker
    - active_worker_agent_run_has_valid_issue
    - role_agents_have_readable_project_task_role_names
    - provider_session_id_is_not_used_as_durable_role_owner
    - active_lease_holder_matches_paperclip_checkout_or_live_run
    - issue_checkout_live_run_agent_registry_and_control_record_agree
    - heartbeat_due_at_not_overdue
    - material_delta_since_last_heartbeat
    - user_questions_are_in_Questions_with_question_first
    - peer_runtime_or_stale_blockers_are_not_misrouted_to_Questions
  allowed_actions:
    - mark_orphan_needs_worker
    - mark_lease_stale
    - mark_blocked_stale_or_looping
    - request_worker_handoff
    - escalate_to_project_orchestrator
    - write_compact_watchdog_digest

Use this for organization health, not ordinary task execution.

scope

Audit the system around the work:

• stale leases;
• orphaned in-progress tasks;
• missing or generic Paperclip issue titles or agent names;
• provider session ids used as worker/reviewer/verifier/watchdog/auditor owners;
• issue lease, checkout/run registry, and control record divergence;
• subagent runs mislabeled as real Paperclip agent runs;
• expired blockers;
• weak evidence;
• tasks closed without verification;
• noisy or missing digests;
• repeated user questions;
• runtime owner conflicts;
• Linear/fallback double-count risk;
• Linear workflow status used as both stage and status;
• Questions polluted by peer/runtime/stale blockers;
• Deferred issues with active leases;
• Reporting or Done before UAT authority is recorded;
• raw Linear issue keys inside machine-readable hidden blocks;
• memory cleanup/update candidates;
• skill/template improvement candidates.

memory policy

Memory should become reusable operational knowledge:

• repeated blockers;
• recurring bad task shapes;
• evidence patterns that worked;
• authority mistakes;
• runtime ownership conflicts;
• digest improvements;
• skill/domain pack improvements.

Do not write noisy memory after every small task. Prefer milestone or audit updates.

degraded tool and bounded stop audit

Record tool degradation instead of hiding it. If a watchdog or auditor cannot use a helper such as context-mode, it should continue with bounded direct reads when safe, write the degraded state, and hand off. If any role loops without material delta, audit whether it wrote a current artifact or explicit gap before stopping.

output format

System health:
Top bottlenecks:
Orphaned work:
Agent/run identity issues:
Issue checkout/live-run/control divergence:
Repeated blockers:
Weak evidence patterns:
Authority risks:
Memory cleanup/update candidates:
Skill/template improvements:
Recommended next actions:

kill switch audit

When kill switch is active, verify no new leases are issued, active workers are stopped or marked orphaned, runtime/public actions are forbidden, emergency digest exists, and recovery options are explicit.