docs: close goal 0041 — OSS trust standards delivered, v0.1.0 live (#82)

* docs: close goal 0041 — OSS trust standards delivered, v0.1.0 live

Acceptance checked against what shipped (all four predicates
evidenced in the archived file); the release exercise's own two
pipeline-defect finds recorded.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01FJ8wStsHyu7XPLTspNjMnQ

* docs: goals 0042-0045 queued — owner-picked post-0041 plan (Group F)

Markdown fidelity pass 2, the overdue hotkey-recorder accelerator
bug, the M365 bridge dry run + capture-path research (absorbing goal
0021 Phase 4's last bullet), and the release-cadence rule with
v0.2.0 gated on 0042.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01FJ8wStsHyu7XPLTspNjMnQ

---------

Co-authored-by: Claude Fable 5 <noreply@anthropic.com>

Author: alicoding

docs/goals/0042-markdown-fidelity-pass-2.md

@@ -0,0 +1,50 @@
+# Goal 0042 — Confluence markdown fidelity, pass 2: close the pinned degradations
+
+Owner-picked 2026-08-13 from the post-0041 planning round. Successor
+to goal 0021 Phase 4's first pass (PR #78), which fixed the
+table-structure loss and pinned five known degradations as goldens in
+`internal/adapters/markdown/testdata/confluence/` with named
+follow-up candidates. This goal implements those five.
+
+## Goal
+
+The five degradations the corpus currently pins become correct
+conversions, each proven by updating its existing golden:
+
+1. **Code-block language hint** — parse Confluence's
+   `data-syntaxhighlighter-params="brush: java; ..."` into a fenced
+   block's language (` ```java `), the shim the reference Python
+   exporter hand-rolls for the same reason (the library only reads
+   `language-*` classes).
+2. **Task-list checkboxes** — `ul.ak-task-list` /
+   `li[data-task-state]` becomes GFM `- [ ]` / `- [x]`.
+3. **Emoji fallback** — `img.emoticon[data-emoji-fallback]` emits the
+   fallback text/emoji instead of today's dead relative image link
+   (currently the worst-case output: worse than dropping).
+4. **Panel type** — info/note/warning/tip panels keep their type in
+   the output (blockquote with a leading bold label, e.g.
+   `> **Warning:** ...` — match what confluence-markdown exporters
+   converge on rather than inventing).
+5. **Expand macro** — `div.expand-container` becomes
+   `<details><summary>` (HTML-in-markdown is valid GFM; the reference
+   exporter does exactly this).
+
+## Plan
+
+Custom rules registered on the adapter's converter (the library's v2
+Register API), Mill-owned code in `internal/adapters/markdown/` —
+commodity library, domain-owned conversion policy, consistent with
+the ports/adapters boundary. Check the library's own extension docs
+for the idiomatic registration shape before writing. Each rule's
+fixture golden updates in the same change; the corpus test is the
+proof. Delegated build once scoped; the five selectors and reference
+behaviors are recorded in goal 0021's Phase 4 research section.
+
+## Acceptance (checkable)
+
+- [ ] All five cases' goldens show the corrected output (no
+      pinned-degradation wording left for them in the test names).
+- [ ] The seven already-correct cases' goldens unchanged.
+- [ ] Full local gate suite green; PR merged green.
+- [ ] Goal 0021's Phase 4 assessment table updated to reflect the new
+      verdicts (this goal referenced from there).

docs/goals/0043-hotkey-recorder-menu-accelerators.md

@@ -0,0 +1,36 @@
+# Goal 0043 — Hotkey recorder vs native menu accelerators
+
+Owner-picked 2026-08-13. The underlying bug is owner-hit and recorded
+in goal 0001's "Live-review additions (2026-08-10)": while the
+Settings hotkey recorder is capturing, pressing a combo that is also
+a native menu accelerator (⌘⇧W closed the window — and Mill exits on
+last-window-close, so the app quit; ⌘Q would too) fires the menu
+action instead of being recorded, because macOS checks menu
+key-equivalents before the webview ever sees keydown. It was deferred
+"after the 0008 build lands" — 0008 landed 2026-08-10, so this is
+overdue.
+
+## Goal
+
+Recording a hotkey can never trigger a native menu action or quit the
+app; permanently-OS-reserved combos are surfaced to the user instead
+of silently failing.
+
+## Plan (from the original live-review note, verify before building)
+
+Suspend the app menu's accelerators while the recorder is active:
+swap in an accelerator-free menu on record-start, restore the real
+menu on capture/cancel/blur — plus a warning in the recorder UI for
+combos macOS reserves at the OS level (which no menu swap can
+reclaim). Needs the real desktop app for verification (menu behavior
+is OS-bound — manual-only registry entry per testing.md, plus unit
+coverage for whatever pure menu-construction logic the fix factors
+out).
+
+## Acceptance (checkable)
+
+- [ ] With the recorder active, ⌘⇧W / ⌘W / ⌘Q are captured as
+      combos, not executed (manual desktop check, registry-listed).
+- [ ] Recorder cancel/blur restores the full menu (manual check).
+- [ ] Any pure logic extracted for the swap is unit-tested.
+- [ ] PR merged green.

docs/goals/0044-m365-bridge-dry-run.md

@@ -0,0 +1,45 @@
+# Goal 0044 — M365 bridge dry run + capture-path research
+
+Owner-picked 2026-08-13. Absorbs goal 0021 Phase 4's remaining
+bullet (the §2.1 M365 bridge dry run) so 0021 can archive on the
+owner's real-use-ready call without carrying open build work.
+
+## Goal
+
+Two deliverables, one session-shaped each:
+
+1. **Bridge dry run** — compose the §2.1 core loop end to end with
+   the pieces that exist today (capture → process → code-exec →
+   clipboard; the "Example: Run copied code" and "Saved page →
+   Markdown" seeds are the starting points), run it against a
+   realistic captured page, and produce a concrete named-gaps list:
+   what's missing between today's Mill and the real
+   copy-from-Confluence → paste-into-M365 daily loop (DOM capture
+   quality, auto-paste target, anything else the dry run surfaces).
+2. **Capture-path research** — the gating unknown from the
+   enterprise-reality context: can a browser extension realistically
+   load in a locked-down enterprise browser (managed-Chrome/Edge
+   extension policy: allowlists, force-install lists, developer-mode
+   blocking — what IT policy typically permits and what precedent
+   exists for getting an internal tool approved), vs. the fallback
+   ladder (save-page capture floor — already shipped; bookmarklet;
+   clipboard-only). Research output is a decision-ready comparison,
+   not a build.
+
+## Plan
+
+Dry run first (it sharpens what the research must answer). Research
+via agent with primary sources (enterprise browser policy docs,
+extension-policy references). No Mill code changes expected beyond
+possibly a seed tweak; if the dry run finds a small fixable gap, it
+rides per the below-goal-granularity rule.
+
+## Acceptance (checkable)
+
+- [ ] The dry run's gap list is written into this file, each gap
+      named with what it would take.
+- [ ] The capture-path comparison (extension vs. fallback ladder,
+      with policy precedent and sources) is written into this file,
+      decision-ready for the owner.
+- [ ] Goal 0021's Phase 4 bullet points here and 0021's archival is
+      unblocked (owner call remains its bar).

docs/goals/0045-release-cadence.md

@@ -0,0 +1,29 @@
+# Goal 0045 — v0.2.0 and release cadence
+
+Owner-picked 2026-08-13, deliberately tiny. v0.1.0 (goal 0041)
+proved the pipeline; this makes releases a rhythm instead of a
+one-off, without adding tooling before the recorded trigger
+(release-please stays deferred until the 2nd+ release exists — see
+archived goal 0041's defer list).
+
+## Goal
+
+A recorded, lightweight cadence rule plus the next release cut by it.
+
+## Plan
+
+Cadence rule (adopt unless the owner objects at cut time): **a
+release tags when a user-visible capability goal archives**, not on a
+clock — the release notes then describe something a user can feel.
+Concretely next: v0.2.0 cuts when goal 0042 (markdown fidelity pass
+2) archives, bundling whatever else landed since v0.1.0 (icon,
+React 19, fidelity pass 1). Mechanics per 0041: bump
+build/config.yml version in a PR, tag the merged commit, release.yml
+does the rest (now with timeouts and the headless-bindings guard).
+
+## Acceptance (checkable)
+
+- [ ] v0.2.0 live on GitHub Releases with attestation verified, cut
+      after 0042 archives.
+- [ ] The cadence rule recorded in SPEC.md's release section (one
+      sentence, LOCKED) in the same change.

docs/goals/BACKLOG.md

@@ -28,6 +28,23 @@ backlog section in the same commit: **work discovered mid-session that
 outlives the session gets a goal file and a queue position before the
 session ends — never left only in an ephemeral session task list.**
 
+**Group F — Real-use fidelity + rhythm (2026-08-13, owner-picked all
+four in the post-0041 planning round; this order recommended and
+accepted)**
+1. [ ] [0042 — Markdown fidelity pass 2](0042-markdown-fidelity-pass-2.md)
+   — implement the five pinned Confluence degradations (language
+   hints, task lists, emoji fallback, panel types, expand→details);
+   corpus + goldens from PR #78 are the proof harness
+2. [ ] [0043 — Hotkey recorder vs menu accelerators](0043-hotkey-recorder-menu-accelerators.md)
+   — overdue owner-hit bug (⌘⇧W while recording closed the app);
+   accelerator-free menu swap during capture
+3. [ ] [0044 — M365 bridge dry run + capture research](0044-m365-bridge-dry-run.md)
+   — absorbs goal 0021 Phase 4's last bullet; gap list + the
+   browser-extension-at-a-locked-down-enterprise decision research
+4. [ ] [0045 — v0.2.0 + release cadence](0045-release-cadence.md)
+   — tag-on-capability-goal-archive rule; v0.2.0 cuts when 0042
+   archives
+
 **Group E — Public-repo hygiene (2026-08-13, owner-directed: source
 comments explain code per standard practice — business/product
 decisions and the owner's own words live in docs/, cited by id, never