Tasks: an opt-in critic review of the locked design

Derived from the approved design.md and testing-plan.md.

Task list

• [x] 1. Teach the compiler the optIn marker

  • Node.opt_in / Node.description in cli/the_loop/graph/model.py, mapped in _build_node and exposed in as_mapping()
  • optIn implies skippable (vocabulary membership + the existing on: skipped edge requirement)
  • refuse required × optIn explicitly, before the implied skippable makes the existing message misleading
  • refuse a skipSets member that is opt_in
  • Depends on: none
  • Requirements: R1.1, R1.2, R1.3
  • Test: T1 — pytest cli/tests/test_graph_skips.py -k "opt_in and compile" (red→green)

• [x] 2. Record a selection in the graph state

  • GraphState.opt_ins (optIns on disk), additive, defaulting to {} on load
  • round-trips through load/as_dict; a pre-issue-188 file loads with no selection
  • Depends on: none
  • Requirements: R1.6, NFR backward compatibility
  • Test: T1/T10 — pytest cli/tests/test_graph_state.py -k opt_in (red→green)

• [x] 3. Default-skip an unselected opt-in node, and say so honestly

  • Runtime.selected() — the mirror filter of declared_skips, honouring an entry only when the compiled graph marks the node optIn
  • declared_skips() folds in every unselected opt-in node as {"via": "not-selected"}
  • _skip_provenance branches on that marker so check reports not selected, never skipped by declaration and never pass
  • _record_selected_skips consumes the gate's optIns, applying the same already-entered and vocabulary guards, and emits graph.opt_ins_selected
  • Depends on: 1, 2
  • Requirements: R1.4, R1.5, R1.6, R1.7
  • Test: T1/T8 — pytest cli/tests/test_graph_skips.py -k "opt_in and (route or report or forged)" (red→green)

• [x] 4. Offer the opt-in phases at phase-selection

  • _phase_rows returns (default_on, opt_in, protected); opt-in nodes are excluded from the default-on rows
  • _checklist_body renders an "Optional phases" section with unticked rows and each node's description
  • _parse_selection returns (skips, opt_ins, refused): ticked opt-in → selected, unticked → not selected, absent → not selected
  • _confirmation names the selected opt-in phases, or says none were selected
  • _frozen_graph carries optIn per node and marks an unselected opt-in node skipped
  • classify_phase_selection returns optIns in its result data
  • Depends on: 1, 3
  • Requirements: R2.1–R2.6, R1.8
  • Test: T1/T8 — pytest cli/tests/test_graph_skips.py -k "opt_in and (checklist or reply or unauthorized)" (red→green)

• [x] 5. Ship the design-critic-review node

  • the node in cli/the_loop/graph/pdlc-work-item-loop.yaml between design and test-planning, optIn: true, stage: critic-review, gating the execution log's Design critic review section; four edges (pass/skipped in, pass/skipped out)
  • the ## Design critic review section in skills/the-loop/templates/execution-log.md (P5c parity)
  • assert the inner PR loop and the contribution loop declare no opt-in node
  • Depends on: 1, 4
  • Requirements: R3.1, R3.2, R3.3, R3.6
  • Test: T2/T12 — pytest cli/tests/test_graph_skips.py -k "opt_in and shipped" cli/tests/test_graph_parity.py (red→green)

• [x] 6. Write the procedure down where a reviewer reads it

  • reference/reviewing.md — the design critic round: subject, prompt contents, where it is recorded, the unavailable rule
  • reference/workflow.md — opt-in phases beside declared skips, and the sequence
  • SKILL.md — one sentence in the selection rule
  • docs/capabilities/process-graph.md + review-loop.md + capabilities.md history rows
  • docs/decisions/decision-071.md + the decisions index
  • Depends on: 5
  • Requirements: R3.4, R3.5
  • Test: T12 — pytest cli/tests/test_docs_parity.py cli/tests/test_writing_parity.py

• [x] 7. Verify: run the plan, capture the evidence

  • every activity in testing-plan.md, ticked only once run; results and evidence committed under evidence/
  • Depends on: 6
  • Requirements: all
  • Test: T1–T13 + full suite

Dependency graph (DAG)

flowchart LR
  T1["1. compiler:<br/>optIn marker"] --> T3["3. runtime:<br/>default-skip + report"]
  T2["2. state:<br/>optIns"] --> T3
  T1 --> T4["4. gate:<br/>checklist + reply"]
  T3 --> T4
  T4 --> T5["5. ship<br/>design-critic-review"]
  T1 --> T5
  T5 --> T6["6. docs, capability,<br/>decision"]
  T6 --> T7["7. verification<br/>+ evidence"]

Checkpoints

After tasks 3, 5 and 6: run the targeted suites named in each task's _Test:_, then make lint typecheck, and append the outcome to execution-log.md. After task 7 the verification node's record is complete and the review chain runs (self → critic → security), followed by the capability/documentation gates and the reviewer briefing.

Review comments

Appended by the-loop's record-feedback hook when a human gate approves with comments.