Skip to content

CHECKLIST.md

Latest commit

 

History

History
52 lines (34 loc) · 2.44 KB

CHECKLIST.md

File metadata and controls

52 lines (34 loc) · 2.44 KB
Raw

TappsMCP checklist (tapps_checklist)

The checklist tracks which MCP tools were invoked in the current checklist session and compares that to task-type policy (required / recommended / optional), adjusted by engagement level and optional .tapps-mcp/checklist-policy.yaml overrides.

Session boundary

  • Calling tapps_session_start starts a new checklist session (new checklist_session_id).
  • The server returns checklist_session_id in session start and checklist responses.
  • Tool calls before the first tapps_session_start in a process use legacy mode: all persisted rows are visible (backward compatible for tests and minimal setups).
  • After a session boundary, only JSONL rows whose session_id matches the active session are evaluated.
  • Use tapps_checklist(..., reset_checklist_session=True) to rotate the checklist session without a full session start (long-lived servers).

Persistence and locking

  • Append log: .tapps-mcp/sessions/checklist_calls.jsonl (each line includes tool_name, timestamp, session_id, success).
  • Active session id file: .tapps-mcp/sessions/checklist_active_session.
  • Writes use a file lock (checklist_calls.jsonl.lock) to reduce corruption under concurrent processes.

Policy version and overrides

  • Responses include checklist_policy_version (hash of merged built-in maps + optional policy file).
  • Optional file: .tapps-mcp/checklist-policy.yaml
extra_required:
  feature:
    - tapps_dependency_scan
extra_recommended:
  review:
    - tapps_validate_changed

Strict task types

  • Set checklist_strict_unknown_task_types: true in .tapps-mcp.yaml (or env TAPPS_MCP_CHECKLIST_STRICT_UNKNOWN_TASK_TYPES) so unknown task_type values error instead of falling back to the review policy.

Outcome-aware mode

  • Set checklist_require_success: true (or TAPPS_MCP_CHECKLIST_REQUIRE_SUCCESS) so the latest receipt per tool must be success: true for that tool to count. Failed gates / scans record failure for key tools when possible.

OpenTelemetry hints (optional)

If the host exports:

  • TAPPS_OTEL_TRACE_ID
  • TAPPS_OTEL_SPAN_ID

checklist responses include otel_trace_hint for correlation with external traces.

Epic markdown validation

Pass epic_file_path to tapps_checklist (with task_type="epic" recommended) to attach epic_validation (structural checks on the epic document).