Tasks: reset the-loop CLI's state for a work item

Phase 3 of 3 (requirements → design → tasks). A DAG of implementation tasks derived from the approved design. MUST be reviewed/approved before implementation begins.

Task list

• [x] 1. SessionRegistry.forget — delete a session record

  • The one primitive reset needs that the registry never had: remove <registryDir>/<slug>.json rather than transition it. False when absent, and an OSError propagates for the caller to collect.
  • Depends on: none
  • Requirements: R1.3
  • Test: pytest cli/tests/test_reset.py -k forget (red→green)

• [x] 2. Dispatcher.close_session reports whether a workspace was removed

  • _cleanup_workspace already computes it and discards it; return it, and return it from close_session. Existing callers ignore the value.
  • Depends on: none
  • Requirements: R1.8, R5.4
  • Test: pytest cli/tests/test_reset.py -k workspace_removal_is_reported (red→green)

• [x] 3. the_loop.reset — the domain

  • ResetOutcome, reset_work_item, work_items_with_state. Close (when live) → forget the record → clear control → clear poll, every step wrapped so an OSError lands in errors instead of raising. Clearing goes through WorkItemStore.write_section(..., None) so the seal-vs-delete rule holds.
  • Depends on: 1, 2
  • Requirements: R1.1, R1.2, R1.4, R1.5, R1.6, R1.7, R2.5, R3.1, R3.2, R3.3, R3.4
  • Test: pytest cli/tests/test_reset.py (red→green)

• [x] 4. --dry-run semantics in the domain

  • Report the same list without closing, deleting or emitting.
  • Depends on: 3
  • Requirements: R4.4
  • Test: pytest cli/tests/test_reset.py -k dry_run (red→green)

• [x] 5. Security: the removal target can only come from a parsed ref

  • No path concatenation in the module; refs reach the stores as a sanitised slug.
  • Depends on: 3
  • Requirements: R2.6, abuse cases 1, 2
  • Test: pytest cli/tests/test_reset_integration.py -k hostile_ref — negative: /, .., a leading - and a null byte are all rejected with nothing removed (red→green)

• [x] 6. The reset sub-parser and its output

  • Selector rules (R2.3/R2.4), all-or-nothing ref validation, per-item reporting, the lazily-built dispatcher-backed close, exit codes 0/1/2.
  • Depends on: 3, 4
  • Requirements: R1.8, R2.1, R2.2, R2.3, R2.4, R2.6
  • Test: pytest cli/tests/test_reset_integration.py (red→green)

• [x] 7. The two warnings

  • A live receiver pidfile (R5.1) and a config that can re-spawn a first-sight item (R5.2), printed once per invocation.
  • Depends on: 6
  • Requirements: R5.1, R5.2, R5.3
  • Test: pytest cli/tests/test_reset_integration.py -k warns (red→green)

• [x] 8. Security: session.reset in the event catalog, and the log stays append-only

  • New EVENT_TYPES entry; a test asserting the log's earlier bytes are untouched by a reset.
  • Depends on: 6
  • Requirements: R4.1, R4.2, R4.3
  • Test: pytest cli/tests/test_reset_integration.py -k event_log — negative: the log cannot be rewritten by the command that could most want to (red→green)

• [x] 9. Security: --all touches only records the stores wrote

  • Depends on: 6
  • Requirements: R2.2, abuse case 3
  • Test: pytest cli/tests/test_reset_integration.py -k stranger — negative (red→green)

• [x] 10. Docs: the command page and the state page

  • docs/cli/commands/sessions.md gains reset; docs/cli/state.md gains how each classified path is erased.
  • Depends on: 6, 7
  • Requirements: R6.1, R6.2
  • Test: pytest cli/tests/test_docs_parity.py, markdownlint-cli2 "**/*.md"

• [x] 11. Capability docs + decision record

  • docs/capabilities/cli.md and interactive-sessions.md (behaviour + history rows), docs/decisions/decision-050.md + index row.
  • Depends on: 10
  • Requirements: R6.3
  • Test: markdownlint-cli2 "**/*.md"

• [x] 12. Full validation

  • make check equivalent: pytest, ruff check/format, pyright, markdownlint, config validation. Record red→green evidence in the execution log.
  • Depends on: all
  • Requirements: R6.4
  • Test: uv run --project cli python -m pytest -q cli and the gates

Dependency graph (DAG)

graph LR
    T1[1 forget] --> T3[3 domain]
    T2[2 close_session bool] --> T3
    T3 --> T4[4 dry-run]
    T3 --> T5[5 sec: refs]
    T3 --> T6[6 sub-parser]
    T4 --> T6
    T6 --> T7[7 warnings]
    T6 --> T8[8 sec: event log]
    T6 --> T9[9 sec: --all]
    T7 --> T10[10 docs]
    T6 --> T10
    T10 --> T11[11 capability + decision]
    T5 --> T12[12 validation]
    T8 --> T12
    T9 --> T12
    T11 --> T12

Checkpoints

• After task 3: the domain suite is green with no CLI wiring — the erasure rules are provable before anything can be typed at them.
• After task 6: argv-in/exit-code-out works end to end.
• After task 9: every negative test from § Security design passes.
• After task 12: the full suite and all gates, recorded in execution-log.md, followed by the self-review rounds and the security-review gate.

Review comments

Appended by the-loop's record-feedback hook when a human gate approves with comments (issue-109). Append-only and attributed: an approval never silently discards a reviewer's suggestions, and the feedback travels with the document it concerns rather than living in a side-channel tracker.