[MakerFlyFly]

Summary

• Add magpie-discourse topic list --category <id|slug|url|slug/.../id> [--since <ISO-8601>] [--limit N] so projects can discover category discussions before reading individual topics. Include subcategories and reuse the existing GET client.
• Request descending activity order, filter on bumped_at and preserve native pinned positions. Look ahead for an extra unique match to prove truncation; pinned topics, including category About topics, count toward the limit.
• Resolve IDs directly and bare slugs only among catalog roots. Keep strict whole-page validation, separate activity and last-message times, and declare Coverage: partial.

Type of change

• Tool / bridge contract
• Python package
• Documentation

Test plan

• Final Linux package suite: 361 tests pass (82 list cases and all 279 unchanged PR1 cases); ruff, format and mypy pass.
• Tests cover native category references, incomplete catalogs, old pinned topics, inclusive cutoffs, deduplication, exact truncation, whole-page errors and fixed parameters on every request. The 3000-topic/two-match regression still uses 2 requests.
• Five integration members pass: vendor-neutrality-score 42, skill-and-tool-validator 493, dev 372, spec-validator 87 and spec-inventory 31.
• All-file non-test hooks: 40 Passed; workspace pytest intentionally separate. Committed-diff pytest selects only Discourse and passes 361 cases.
• Magpie complete-PR A–E checklist passes for all three commits, with 0 blocking / 0 advisory items.
• zizmor 1.30.1 offline: no unsuppressed findings (1 ignored, 18 suppressed).
• Final anonymous CLI verification: 4 actual HTTPS GETs; 5 including the planning probe, below the round limit of 10. No forum writes.
• Native Claude validation is internally skipped because its CLI is absent. Test collections overlap and are not added as independent totals; full-workspace pytest was not run.
• Cloud checks are separate: prek was cancelled for both 7f5d83d0 and 4e72eb9. Neither is counted as passed; current-head results are assessed separately.

RFC-AI-0004 compliance

• Sandbox — forum origin/path restrictions remain; no sandbox or host-allowance changes.
• Vendor neutrality — Discourse-specific behavior stays in the partial adapter.
• Write-access discipline — GET only, including redirects; automated HTTP tests use mocks.
• Privacy LLM — no model calls; callers establish visibility and apply existing privacy approval before sending content to a model.

Linked issues

Owner audit: MakerFlyFly/magpie#23.
Refs apache/magpie#307, following merged apache/magpie#1755.
This slice does not close the broader request.

Notes for reviewers

• Fixed audit base: b9e14009367f37efce807e473b95c331286036a1 (audit/discourse-topic-list-base). Head: a3794eff6acfb9e9a2635da28fa3632ac65fe8ab (feat/discourse-topic-list). The complete PR contains PR2 only: 10 files, +1018/-9.
• Business-review delta: six files, +278/-476. History is preserved.
• The earlier draft's ID-less parent/child syntax is removed. Use an ID, a same-forum category URL or slug/.../ID for descendants; top-level slug lookup can be incomplete and suggests an ID when unresolved.
• Continuation strings are signals only. Their URL and parameters are not consumed; each request is rebuilt locally with fixed category/activity parameters and an incrementing page. Empty/repeated complete pages claiming continuation still fail.
• Without --since, there is no fixed recent-time window. Filtering returned pinned topics does not refill the result. With since, only an older topic explicitly marked pinned: false establishes a cutoff.
• All received records are validated, including records beyond the output limit or cutoff. The changing list is not a consistent snapshot.
• General resource budgets, retries, Hot/Top ranking, search, writes, shared-client refactoring and remaining archive verbs are outside this slice. Budget design remains after PR2 and before PR3.
• Owner manual review is pending before any upstream submission.

AI disclosure

AI-generated draft for owner audit. Codex (GPT-6) assisted implementation, review-response analysis, tests and validation.
This draft does not claim completed contributor manual review.