chore(ai): archive session record and wrap-up lint followups

Records today's task-review spec + google-keep MCP install work. Lint-followups carries one misplaced-heading judgment (line 2153, false positive — `**` inside `=...=` verbatim) and a 9-entry date-coverage list, both deferred per the task-review spec's plan to retire the date-coverage scan once implemented.

2 files changed, 226 insertions, 0 deletions

diff --git a/.ai/sessions/2026-05-16-12-35-task-review-spec-and-google-keep-mcp.org b/.ai/sessions/2026-05-16-12-35-task-review-spec-and-google-keep-mcp.org
new file mode 100644
index 0000000..67ad58c
--- /dev/null
+++ b/.ai/sessions/2026-05-16-12-35-task-review-spec-and-google-keep-mcp.org
@@ -0,0 +1,211 @@
+#+TITLE: Session Context — inbox lint follow-ups + date-coverage walk
+#+AUTHOR: Craig Jennings & Claude
+#+DATE: 2026-05-16
+
+* Summary
+
+** Active Goal
+
+Three arcs. (1) Absorb the morning's inbox handoff containing seven lint warnings plus eight date-coverage entries the wrap-it-up scan flagged on todo.org. (2) =/brainstorm= a recurring task-review habit so the date-coverage scan can retire — the scan was generating cleanup work because it was solving the wrong problem (assuming high-priority tasks need dates). (3) Install a Google Keep MCP server on both machines via the existing =mcp/= pipeline.
+
+** Decisions
+
+- *Defer the 8 date-coverage entries* to the first task-review cycle rather than walking them tonight — they'll be absorbed when the habit lands. Inbox handoff deleted; its other payloads (cj-comment lint suppression, line 2139 false positive) were either already fixed in =3fb4c80= or inspected and dismissed.
+- *Brainstorm output: pure Emacs custom agenda + keystroke macros* (Approach B) over five other candidates. Rejected: Claude-driven command (heavy daily friction), hybrid Emacs+Claude (premature for v1), random sample (no coverage guarantee), write-time hook (silent on quiet weeks), section-rotating (uneven section sizes).
+- *State model:* =:LAST_REVIEWED:= property drawer per top-level task; sort NIL-first so untouched tasks surface naturally; batch N=7 (~12-day cycle on ~80 open tasks).
+- *Promote =task-review.org= to template-level* (canonical at =claude-templates/.ai/workflows/=, rsync'd to every project), with three trigger aliases in =INDEX.org=. Startup nudge stays project-specific via =startup-extras.org=. Wrap-up watchdog flips from "do all priorities have dates?" to "is the review habit happening?" at a 30-day threshold.
+- *Google Keep MCP server: feuerdev/keep-mcp* via =uvx= over kud/mcp-google-keep (macOS Keychain) and davenicoll. Linux-friendly, env-var auth fits =secrets.env.gpg=, default-safe modification scope (only modifies notes carrying its =keep-mcp= label).
+- *Master token retrieval: path #2 (DevTools EmbeddedSetup + gpsoauth.exchange_token).* Path #1 (password-based gpsoauth.perform_master_login) is restricted by Google since 2025.
+- *Invocation workaround: =uvx --from keep-mcp python -m server.cli=.* The README's =pipx run keep-mcp= form doesn't translate cleanly to =uvx= — keep-mcp's =mcp= entry-point script collides with the official =mcp= SDK's same-named script (a transitive dep). Module-level invocation bypasses the collision.
+
+** Data Collected / Findings
+
+- =/lint-org --check= on =todo.org= post-=3fb4c80=: 0 mechanical, 1 judgment remaining (line 2139, =misplaced-heading=). The =3fb4c80= cj-comment suppression eliminated the six false positives at lines 16 and 1291.
+- Line 2139 misplaced-heading is itself a false positive: =**= inside =...== verbatim markers (correctly wrapped). Leave alone.
+- The wrap-it-up date-coverage scan is read-only — it doesn't modify =todo.org=, just appends candidates to the lint-followups file. The recurring noise comes from candidates piling up across wrap-ups without per-entry disposition.
+- todo.org has ~80 open top-level =[#A]= / =[#B]= / =[#C]= tasks (children inherit parent priority and don't count as separate review units).
+- keep-mcp package is on PyPI as =keep-mcp= 0.3.1 (= feuerdev/keep-mcp). Package's =[project.scripts]= declares =mcp = "server.cli:main"=; the module is literally =server= (not =keep_mcp=).
+- MCP =initialize= handshake against =uvx --from keep-mcp python -m server.cli= returns =keep v1.27.1=, confirming the right entry point boots.
+- =claude mcp list= on both machines reports =google-keep: ... ✓ Connected=.
+- =ssh velox= from ratio resolves to tailscale (=100.127.238.103= via =velox.tailf3bb8c.ts.net=) but timed out from this session. =velox.local= via mDNS works once the host is up.
+- velox needs explicit =PATH=$HOME/.local/bin:$PATH= prefix for non-interactive ssh — =uv= and =claude= both live there and aren't picked up by default.
+
+** Files Modified
+
+Three commits pushed to =origin/main=:
+
+- =70af2a0 chore(ai): sync lint-org and wrap-it-up from claude-templates= — Phase A startup picked up =3fb4c80= (cj-comment lint suppression) and =684891d= (LINT_ORG_FOLLOWUPS default to =./inbox/=) that hadn't propagated to the project mirror.
+- =9c5aff1 feat(mcp): add google-keep server= — =mcp/servers.json= entry plus =mcp/secrets.env.gpg= reground with two new env vars. Body documents the entry-point-collision workaround for future archaeology.
+- =daf960f docs(design): task-review spec + filed [#A] task= — new =docs/design/task-review.org= (~350 lines: problem, six approaches, recommended design, six chunks, open questions, six-step Next Steps). =todo.org= gains a top-of-section =[#A]= entry pointing at the spec.
+
+Working tree at wrap time: just =.ai/session-context.org= (becomes the session record via rename).
+
+** Next Steps
+
+- *Smoke test the keep MCP* after a Claude Code reload on either machine — Craig will return for this.
+- *Implement the task-review spec.* Six components in order: (1) extract =task-review-staleness.sh= with bats tests, (2) replace date-coverage section in =wrap-it-up.org= with the health-check, (3) author =task-review.el= in archsetup with ERT tests, (4) author =task-review.org= workflow + INDEX entry, (5) add startup nudge to =.ai/project-workflows/startup-extras.org=, (6) smoke test.
+- *Carryover from prior sessions remains:* =[#C]= consolidate =claude-templates/Makefile= post-fold, =[#C]= wrap-it-up Step 3.5 GitHub-family-remote assumption, =[#B]= Phase A inbox-scan blind to =claude-templates/inbox/=, =DOING [#A]= memory-sync (pending VERIFY on stow approach), =[#A]= =/update-skills= skill, =[#A]= =create-documentation= skill, =[#A]= 2026-05-04 audit review pass.
+
+* Session Log
+
+** Startup (08:58 CDT)
+
+Clean startup — no =session-context.org=, no cross-agent traffic, no Active
+Reminders, no Pending Decisions. Template syncs ran clean. Read the 5 most
+recent Summaries. Inbox had one file:
+=2026-05-16-0512-from-work-lint-followups-handoff.org= — a re-routed payload
+from the work project containing (a) seven lint warnings against
+=rulesets/todo.org= and (b) eight =[#A]=/=[#B]= tasks without
+=DEADLINE=/=SCHEDULED= flagged by wrap-it-up's date-coverage scan.
+
+Noted that two of the root causes the handoff calls out are already fixed:
+=3fb4c80 feat(lint-org): recognize cj-comment blocks= (suppresses the
+six line-16/line-1291 false positives) and =684891d fix(workflows):
+default LINT_ORG_FOLLOWUPS to current project inbox= (the routing bug that
+sent these to the wrong project's inbox in the first place).
+
+Craig picked option 1: re-run =/lint-org --check= to confirm the cj-comment
+suppression, then walk line-2139 and the date-coverage list together.
+
+** Confirmed cj-comment suppression; line 2139 inspected
+
+=lint-org --check= on =todo.org=: 0 would-fix mechanical, 1 judgment
+remaining (line 2139, =misplaced-heading=). All six cj-comment warnings
+gone — =3fb4c80= confirmed working.
+
+Line 2139 is inside the body of a =** DONE [#B] Document startup
+pull-ordering rule= entry. The text is =before proceeding" rule shipped
+2026-05-15 in the =** Startup Pull Ordering== — the =**= sits inside
+=...== verbatim markers and is correctly wrapped. False positive, leave
+alone.
+
+** Pivot: /brainstorm on task-review habit (09:something CDT)
+
+Walked the eight date-coverage entries up to row 1; surfaced two
+encoding-vs-design-intent tradeoffs (durable inline =# no-date by
+intent= comment vs. closing-paragraph rationale that the operator
+dismisses no-date in daily-prep, not in =todo.org=). Then read the awk
+scan together and confirmed it doesn't modify =todo.org=, just appends
+flagged candidates to =$followups=.
+
+Craig reframed: the whole org-lint date-coverage business is symptom of
+a wrong assumption — that =[#A]=/=[#B]= tasks /need/ dates. They don't.
+Some tasks are important without being time-bound. What's actually
+needed is a regular habit of walking the task list, reverifying that
+each task is still important, well-written, and actionable. With that
+habit in place, the date-coverage scan becomes redundant and can be
+retired.
+
+Invoked =/brainstorm=. Phase 1 question 1 asks about the failure mode
+the habit should prevent (wrong priorities / vague entries / forgotten
+entries / stale entries / all of the above).
+
+** Brainstorm output — task-review design
+
+Phase 1 close: /A daily 5-min review surfacing N oldest-unreviewed
+top-level =[#A]= / =[#B]= / =[#C]= tasks for re-grade, kill, or light
+prose touch — rotating through the list over weeks — until Craig
+trusts the priorities and content of =todo.org= at a glance./
+Failure modes addressed in order: priority drift, stalled commitments,
+prose decay. Success metric: trust (subjective), not throughput.
+
+Phase 2: 6 approaches considered (3 conventional, 3 tail). Picked
+Approach B — pure Emacs custom agenda + keystroke macros. Rejected
+Claude command (too heavy daily), hybrid (premature for v1), random
+sample (no coverage guarantee), write-time hook (quiet weeks miss),
+section-rotating (uneven section sizes).
+
+Phase 3 — design chunks walked + confirmed:
+1. State model: =:LAST_REVIEWED:= property drawer per top-level task.
+2. Architecture: single elisp file =task-review.el= in archsetup.
+3. Selection: top-level =**=, open keywords, =[#A]=/=[#B]=/=[#C]=,
+   sort by =LAST_REVIEWED= asc, NIL first. Batch N=7 (~12-day cycle).
+4. Keymap: =1=/=2=/=3=/=r=/=k=/=d=/=e=/=s=/=q=. Confirm on kill.
+5. Retire date-coverage scan; replace with review-habit health check
+   in wrap-up. Threshold 30 days. Watchdog only — one line if stale.
+6. Workflow promoted to template-level. =task-review.org= in
+   =claude-templates/.ai/workflows/=. INDEX trigger aliases:
+   =task review=, =review tasks=, =task-review=. Startup nudge
+   stays project-specific via =startup-extras.org=. Threshold 7 days.
+7. Testing: ~20 ERT (data layer + actions), ~5 bats (staleness
+   script), manual smoke checklist. Extract staleness math to
+   =task-review-staleness.sh= as single source of truth.
+
+Design doc written to =docs/design/task-review.org= (used =.org= per
+file-format preference, not =.md= as the brainstorm skill's default).
+
+Filed implementation task at =todo.org:10= → =** TODO [#A] Implement
+task-review daily-habit per spec=, body links the spec + lists the
+six implementation components.
+
+** Cleanup
+
+Inbox handoff =2026-05-16-0512-from-work-lint-followups-handoff.org=
+deleted — signal fully absorbed: (a) cj-comment lint suppression
+confirmed working post-=3fb4c80=; (b) line-2139 misplaced-heading
+inspected as false positive (=**= inside =...== verbatim markers,
+correctly wrapped, leave alone); (c) eight date-coverage entries
+deferred — they'll be absorbed into the first task-review cycle once
+the spec is implemented. The replacement health-check (threshold 30
+days at wrap-up, 7 days at startup) takes over the watchdog role.
+
+** Pivot: install Google Keep MCP server
+
+Brainstorm artifacts (=docs/design/task-review.org=, =todo.org=
+=[#A]=, =.ai/session-context.org= log) all on disk. Working tree
+has uncommitted changes — implementation deferred to a future
+session via the filed task; brainstorm artifacts will commit at
+wrap-up. Pivoting to new ask: install a Google Keep MCP server
+into the existing =mcp/= pipeline.
+
+** Google Keep MCP install — ratio
+
+Picked =feuerdev/keep-mcp= (PyPI =keep-mcp= 0.3.1) over kud or
+davenicoll. Reasons: =uvx= analogue of existing =npx -y= pattern,
+env-var auth fits =secrets.env.gpg=, default-safe modification scope
+(only modifies notes carrying its =keep-mcp= label), Linux-friendly.
+
+Master token retrieved via DevTools EmbeddedSetup flow (path #2 —
+=accounts.google.com/EmbeddedSetup= → =oauth_token= cookie →
+=gpsoauth.exchange_token=). Token stored in =mcp/secrets.env.gpg=
+as =GOOGLE_KEEP_MASTER_TOKEN=, with =GOOGLE_KEEP_EMAIL= set to
+=craigmartinjennings@gmail.com=. Re-encrypted via symmetric
+=gpg -c --cipher-algo AES256=, plaintext lived only in =/dev/shm=
+and was =shred=-deleted.
+
+Invocation gotcha: =uvx --from keep-mcp mcp= resolved to the
+official =mcp= SDK CLI (shipped as a dependency of =keep-mcp= and
+shadowing keep-mcp's own =mcp= entry-point script). Verified via
+=server.cli= module-invoke: =uvx --from keep-mcp python -m server.cli=
+boots cleanly and answers an MCP =initialize= handshake as =keep
+v1.27.1=. Final =servers.json= entry uses that =python -m server.cli=
+form, not the README's =pipx run keep-mcp= form.
+
+Working tree state at this point:
+- =mcp/servers.json= — modified (new =google-keep= entry).
+- =mcp/secrets.env.gpg= — modified (new ciphertext after symmetric
+  re-encrypt with the two added env vars).
+- =mcp/gcp-oauth.keys.json= — touched by =install.py= (pre-existing,
+  decrypted by install regardless).
+
+=claude mcp list= verified: =google-keep: uvx --from keep-mcp python
+-m server.cli - ✓ Connected=.
+
+Velox install pending: needs the new commits pushed to =origin/main=,
+then ssh + pull + =python mcp/install.py= on velox to register
+locally there. Velox already has =uv= per recent activity.
+
+** Commits + push
+
+Three commits landed on =origin/main=:
+- =70af2a0 chore(ai): sync lint-org and wrap-it-up from claude-templates= — Phase A startup brought in the canonical =3fb4c80= (cj-comment lint suppression) and =684891d= (LINT_ORG_FOLLOWUPS default fix) deltas that hadn't yet propagated to the project mirror.
+- =9c5aff1 feat(mcp): add google-keep server= — adds =google-keep= to =servers.json= via =uvx --from keep-mcp python -m server.cli=, with the entry-point-collision rationale in the body. =secrets.env.gpg= grew two vars.
+- =daf960f docs(design): task-review spec + filed [#A] task= — the brainstorm output, plus the implementation task at =todo.org:10=.
+
+Pre-push reconcile 0/4, push clean.
+
+** Velox install
+
+First attempt failed because non-interactive ssh has no DISPLAY and no TTY — gpg pinentry-gtk silently cancelled. Retried with =ssh -t velox.local= and =GPG_TTY=$(tty)=. The "pseudo-terminal will not be allocated" warning fired (Bash tool subprocess has no TTY to allocate) but pinentry did pop a dialog on velox — Craig didn't notice it for a moment, then answered. Install ran cleanly; =claude mcp list= verified =google-keep: ... ✓ Connected=.
+
+Reaching velox: =ssh velox= timed out (resolved to tailscale IP =100.127.238.103= via =velox.tailf3bb8c.ts.net=, but that route was unreachable from this session). =velox.local= via mDNS worked once Craig confirmed the host was up.
diff --git a/inbox/lint-followups.org b/inbox/lint-followups.org
new file mode 100644
index 0000000..b87722a
--- /dev/null
+++ b/inbox/lint-followups.org
@@ -0,0 +1,15 @@
+
+* 2026-05-16 lint-org follow-ups — todo.org
+** TODO line 2153 — misplaced-heading — Possibly misplaced heading line
+
+* 2026-05-16 Sat — Date coverage: [#A] / [#B] tasks without DEADLINE or SCHEDULED
+Review each: add a date, drop the priority, or confirm 'no-date by intent' inline.
+- 10: ** TODO [#A] Implement task-review daily-habit per spec
+- 28: ** DOING [#A] Check that memories are sync'd across machines via git.m
+- 58: ** TODO [#A] Build =create-documentation= skill for high-quality project/product docs
+- 692: ** TODO [#A] Review pass: tighten skills and rulesets after 2026-05-04 audit
+- 1221: ** TODO [#B] Build =ov-1= skill for DoDAF OV-1 (High-Level Operational Concept Graphic)
+- 1295: ** TODO [#A] Build =/update-skills= skill for keeping forks in sync with upstream
+- 1621: ** TODO [#B] Add =make remove= for interactive ruleset removal via fzf
+- 1686: ** TODO [#B] Document the =mcp/= install pipeline in =mcp/README.org=
+- 1773: ** TODO [#B] Phase A startup blind to =claude-templates/inbox/= post-fold :bug:fold: