Tasks: the registry lists the files it wrote (issue-111)

Phase 3 of 3. Derived from the locked design.md. TDD (tdd.mode: standard): the failing test comes first for every behavioural task.

DAG

flowchart LR
    T1[T1 regression test:<br/>poll-state neighbour] --> T3[T3 name-aware scan]
    T2[T2 corruption + slug<br/>coverage tests] --> T3
    T3 --> T4[T4 full gate]
    T4 --> T5[T5 docs: capabilities]
    T5 --> T6[T6 PR + reviewer briefing]

Task list

• [x] T1 — Failing regression test (cli/tests/test_routing.py, beside the existing registry cases): a registry directory holding one registered session and a poll-state.json lists exactly that session and emits no warning (caplog.at_level(logging.WARNING)). Fails before T3 on the log assertion. (AC2, AC3, AC6)
• [x] T2 — Corruption + slug-coverage tests: (a) rename the existing garbage.json case to a registry-shaped filename so it keeps exercising the corruption path, and assert the warning is logged — plus a registry-shaped file that is valid JSON but has no workItem; (b) a test that registers every ref shape the registry can write (multi-digit numbers, dots and hyphens in owner/repo, characters slug sanitises) and asserts each one lists, with the expectation driven from WorkItemRef.slug rather than literal names. (AC1, AC4)
• [x] T3 — Name-aware scan in cli/the_loop/sessions/registry.py: module-level _REGISTRY_FILE_RE (anchored fullmatch, documented as a superset of what _write produces and why the directory is shared), and a debug-level skip in list_sessions before _read. No other function changes. (AC1, AC2, AC3, AC5)
• [x] T4 — Full gate from the repo root (make check): ruff check, markdownlint, ruff format --check, pyright, validate_config, pytest. Record the red→green evidence in the execution log. (AC6)
• [x] T5 — Documentation: docs/capabilities/cli.md and docs/capabilities/webhook-triggers.md — the <root>/sessions/ directory is shared session-related state and the registry reads only the files it wrote, with the corruption warning reserved for those; history rows added. (AC7)
• [x] T6 — PR + reviewer briefing from skills/the-loop/templates/pr-briefing.md; execution log finalised; label → loop:needs-review. (ready-to-ship gate)

Verification

Each behavioural task's evidence is its test command's red→green transition, recorded in execution-log.md. The work item is ready to ship when the full gate is green, the capability docs are updated in this same PR, the security review is recorded, and the briefing is posted.