Archive — Done Log

Append-only, out-of-context record of retired done-log items. Never pruned, reviewed, or nagged. Entries are dated prose bullets, each the original + **Archived YYYY-MM-DD** — reason.

Age-drained 2026-06-09

(Moved from current-plan.md's Done log — granular blocks older than 7 days, beyond the 5-most-recent floor. Phase-grouped, verbatim. First age-drain of the /hygiene-check build.)

Done 2026-05-29:

  • Reframed Yggdrasil as the plan/design partner; moved brainstorming in from the personal layer and groomed it (structure, tone, frontmatter, output path). (Remaining brainstorming-framework rework is bookmarked, no longer a Phase 4 line item.)
  • Built and wired the /save-progress command; recorded the self-housekeeping auto-commit exception in CLAUDE.md and AGENTS.md; later refined it to own the next-session pointer, reconcile the plan body, and enforce single-source-of-truth metadata (de-duplicated current-plan.md's header).
  • Added the ## Design Philosophy section to the personal CLAUDE.md (low pain tolerance + complexity-cost notes, Otwell-derived).

Done 2026-05-28:

  • Added AGENTS.md to the Yggdrasil repo root, closing the long-standing bookmark and matching the "AGENTS.md at every repo root" decision.
  • Made the personal layer its own git repo, then moved it into a dedicated C:\Seafile\Claude\personal\ subdirectory — so the library root is three clean sibling repos (personal/, yggdrasil/, odin-codin/) with none nested inside another. Re-pointed the ~/.claude/ links and updated the sync docs (CLAUDE.md, new-machine-setup).
  • Captured a hard-won link lesson: PowerShell 5.1's New-Item -SymbolicLink ignores Developer Mode and demands admin; cmd's mklink honors Dev Mode without elevation, and junctions need no privilege at all. Folded into new-machine-setup.
  • Realigned ~/.claude/skills to per-skill aggregation (a real dir with one junction per skill from each repo's own skills/), matching the catalog's "subdirectory-level symlinks" design. Added relink.ps1 — idempotent, admin-free — as the single source of truth for wiring ~/.claude/.
  • Reviewed laravel-conventions (personally-authored, just .env hygiene) and renamed it project-conventions to match its general scope; re-linked.
  • Removed the throwaway new-machine-setup-workspace/ eval output from the personal skills dir.
  • Nuked the personal skill-creator and mcp-builder copies — both redundant with enabled official plugins (skill-creator@claude-plugins-official, mcp-server-dev@claude-plugins-official). The plugins are Anthropic-maintained and auto-updating; mcp-server-dev supersedes mcp-builder with build-mcp-server/build-mcpb/build-mcp-app. (Personal skill-creator was an older, un-customized snapshot, not byte-identical to the plugin; Seafile retains it.)
  • Fixed the Stop-hook quality-gate false-positive that was deadlocking turns (see session lessons).

Age-drained 2026-06-16

(Age-drained from the current-plan.md Done log — granular blocks older than 7 days, beyond the 5-most-recent floor. Verbatim. Second /hygiene-check age-drain.)

Done 2026-06-08 (planning skill — the flagship — built):

  • The planning skill is built and wired in. Authored skills/planning/SKILL.md (~373 lines) via a section-by-section wording walk against superpowers' writing-plans — frontmatter (disable-model-invocation: true, allowed-tools: [Read, Glob, Grep]) through Exit Criteria + Key Principles. Carries: the verifiable-outcome task anchor (Done-when/Confirm-by; machine checks as evidence-for-the-human); the ## Deferred No-Placeholders carve-out (masquerade-only enforcement); git-by-default worktree isolation; the optional 3-way lens audit (skip / single-thread / subagents) with the four orthogonal lenses; a human-in-the-loop execution handoff (no autonomous fork). Wired via relink.ps1 (junction verified); discoverable as /planning next session. Closes the headline "Next build" and the coupled brainstorming-review effort — the brainstorming→planning seam is now closed both directions.
  • Design consolidated. The previously-distributed design (this doc's "Planning-skill design" / "Next build" / catalog + list-hygiene §5.6) extracted into a durable working/2026-06-08-planning-skill-design.md. Archival location once feature-complete is deferred (Brad).
  • Coherence pass: trimmed superpowers-provenance from the operative text, reordered Map→Header→Tasks, deduped the sizing statement. Docs reconciled (AGENTS.md / README.md skill listings; the deeper AGENTS/README methodology mention parked [priority]). Two bookmarks captured (checkbox tracking; git-by-default in the bookend wiring).

Done 2026-06-08 (tracked-deferral settled; list-hygiene extended; settings decision):

  • Planning-skill §10 tracked-deferral — SETTLED. Designed it as the plan-deferral descendant (working/2026-06-05-list-hygiene-design.md §5.6): a ## Deferred section in the plan doc, dated-prose bullets with an encouraged-not-required revival trigger (soft triggers first-class); placement in the section — not the trigger's quality — is what makes an item tracked; No-Placeholders enforcement catches only a TBD masquerading as an active task. Closes the last open planning-skill design question — wording walk is now unblocked.
  • List-hygiene design extended: bookmarks reclassified out-of-context, on-demand (§5.1/§8, ledger-surfaced; supersedes the 2026-06-03 single-doc call); new per-item aging anti-rot primitive (§4 #8, ~14d uniform, informational, never forced); the heat ↔ context-load principle named. Five new decision-log entries.
  • Settings.json continuous-sync — decided against (won't build); personal-layer settings hand-managed (new personal/CLAUDE.md "Settings sync" convention; bookmark shrunk to residual; claude-config/README.md + plan Hook-infra prose reconciled).
  • "Design without a demand character" (autonomy/PDA + spoons) captured as a standing personal/CLAUDE.md Design-Philosophy principle + a 2026-06-08 session lesson. personal/bookmarks.md stood up (first personal-scoped store) with the exportable communication-context-model idea.

Done 2026-06-06 (puppet/scry first-run test + refinements):

  • First real /puppet + /scry run (test) against the FFXI RE project (C:\Projects\XI-RE-VibeTesting). Built the onboarding cache (PUPPET.md), wore the kit (9 emulated skills), verified most of the 8-point plan — full scorecard in the "Foreign-project adoption" section. Finding #1 resolved: writes work via the .puppet/** whitelist; allowed-tools doesn't constrain.
  • /scry test: cloned XiView (KenshiDRK) + Caradog-XI-View from source into research/repos/; self-assessment passed. Refinements: scry URL-discovery rule, puppet "don't stress" rule, WebSearch whitelisted (personal layer). New findings bookmarked (clone hygiene; gh-on-PATH + new-machine-setup STOP pattern). Added puppet + scry to AGENTS.md's commands row (consistency-check fix).
  • Still to run: #1 + the writes-half of #2 closed 2026-06-06 via a fresh janet target (zero permission prompts, Brad-confirmed — see the "Second first-run test" note + scorecard). #8 (/scry no-target warning) closed 2026-06-06 via a no-target gate test (gate fired, halted clean). #2 clone-half closed 2026-06-07 via a live /scry clone against the adopted janet target — watched git clone auto-approved with zero prompts (Brad-confirmed). Puppet/scry first-run verification now fully complete. (#6 skill-emulation ✅ and #3 cache-hit ✅ 2026-06-06 — see scorecard + checkpoint.)

Done 2026-06-07 (puppet/scry #2 clone-half closed + scry atomic-clone rule):

  • Ran /puppet C:\Projects\janet (cache hit) → /scry to research autism/ADHD assistant skillsets. A watched git clone (into research/repos/) auto-approved with zero prompts (Brad-confirmed), closing the last open scorecard item, #2 clone-half — puppet/scry first-run verification is now complete.
  • Refinement (commands/scry.md): an earlier clone attempt prompted because it was chained (cd && mkdir && git clone …) — the Bash(git clone:*) permission is a prefix rule and only matches a command line starting with git clone. Codified the clone step as a single bare git clone <url> <absolute-dest>, never chained with cd/mkdir/&& (git creates parent dirs on its own). Also re-confirmed gh is not whitelisted (it prompts; only git clone is) — consistent with the existing gh-on-PATH bookmark.
  • Follow-up session (2026-06-06): completed #6 by emulating ffxi-status-info for a full 3-way 57.DAT compare (Backup/Normal/Widescreen) against the scry'd XiView repo + a hash mislabel-check (all clean); results in PUPPET.md. #3 cache-hit confirmed at this session's startup. The 1081 "dubious" variant was resolved as a corrupted source file (re-test with better files), and the reconciling fix was applied to the target's ffxi-status-info/SKILL.md (saved to the read-only repo, not committed there). Scorecard #6 update committed a2e594e.
  • Earlier same day (committed 3be536f): the in-flight-design durability guardrail (/save-progress step 1(a)) + bookmark, surfaced chasing a puppet design doc that never existed (confirmed across git, disk, Seafile log + server trash).

Done 2026-06-06 (puppet/scry availability fix + design-intent capture):

  • Sat down via /good-morning; found /puppet + /scry unavailable — committed in 96153c1 but never symlinked because relink.ps1 wasn't re-run after that session. Fixed with relink.ps1 (both now linked).
  • Audited the pair (built-but-never-run); verified across all git history that no design doc ever existed. Recorded the design intents in the new "Foreign-project adoption — /puppet + /scry" section, with Brad's 8-point first-run test plan preserved there. Reframed puppet as a multipurpose foreign-toolkit sandbox (incl. community/public skill sets), with the RE/DAT work as a borrowed legacy instance — explicitly not Odin Codin' (server-side).
  • Bookmarked: [priority] good-morning-reuse bug; a fresh-eyes zero-context subagent review pass; label-encoding review; a proper puppet design review for later; and the design-doc-organization scheme. Corrected the Phase 10 DAT-file conflation.

Done 2026-06-05 (accumulating-list hygiene design):

  • Designed the accumulating-list hygiene system end-to-end — the full scope of the spun-off parked-items / bookmark-unification thread, expanded to subsume session lessons, the Done log, the design catalog, and the archive idea. One type system (List/Target/Archive/Scratchpad; axes Scope·State·context-load), one base format (dated prose bullet + optional kit), shared primitives, a per-scope .meta/ledger.yaml, per-source archive/ files, and /bookmarking evolved to capture-only. Written up + reviewed section-by-section, committed to working/2026-06-05-list-hygiene-design.md (4306794). Build deferred to follow-ons (the /hygiene-check command; the §9 migrations).

Done 2026-06-04 (session bookends + hygiene):

  • Consolidated bookmark sprawl (24→20 items): bookends/doc-hygiene cluster, parallel-execution/concurrent-sessions cluster (incl. captured worktree research), creating-workflow-skills merge.
  • Cleaned the deferred /consistency-check loose thread (scratchpad #3/#4).
  • Researched multi-agent-on-one-repo best practice (worktree-per-agent).
  • Designed the git working-tree discipline frame (un-fusing, soft-pin lens snapshots, writes←instances/reads←subagents, adopt-isolation/reject-autonomy, ~2–3 ceiling, bookend-wiring shape) — see the new in-progress section. Read the four superpowers worktree/parallel skills from source + a 2026 web pass.
  • Resolved the Seafile-vs-worktrees question; applied seafile-ignore.txt (library root) and bookmarked phasing Seafile out for GitHub.

Done 2026-06-03:

  • Doc-sync infrastructure. Built the .meta/durable-docs.md manifest and wired both session bookends to it (shared single source of truth; "bookend sibling" cross-refs). Cut the planned /recap command — triage now lives in the bookends + ad-hoc asking + proactive flagging; removed /recap references from the docs (conversational uses kept).
  • Read whitelists. Path-scoped Read(//C/Projects/**) + Read(//C/Seafile/Claude/**) added to the live settings.json, the personal/claude-config snapshot, and the new-machine-setup reconcile list.
  • Personal CLAUDE.md: added the "spin off heavy tangents" Working-Style note.
  • Bookmarks: bumped settings continuous-sync to "soon"; captured additionalDirectories, generic-schema-as-upper-layer, the web-research-phase idea, and the "no Option A/B labels in notes" hygiene note; pruned the resolved source-fidelity entry.

Done 2026-06-02:

  • Resolved the upsert review: demoted the skill to a 2–3 line "upsert" convention in the personal CLAUDE.md (lighter always-on weight than the skill's ever-loaded description, and a standing instruction triggers more reliably than skill auto-fire). Deleted the upsert skill folder, removed its dangling ~/.claude/skills junction by hand, and re-ran relink.ps1. Fixed a CLAUDE.md drift bug found en route — the Seafile-sync section still described ~/.claude/skills as a whole-folder link; corrected it to the real per-skill-junction aggregation. Bookmark hygiene: pruned the resolved upsert and completed doc-sweep items, trimmed the working-copy-convention bookmark to its remnants.
  • Source-fidelity convention + GitHub backstop. Added a "Source fidelity" section to the personal CLAUDE.md: a judgment-gated rule to pause and surface on partial/lossy reads, with the ⚠️ PARTIAL SOURCE signal (Claude→Brad) and the canonical / from source keyword (Brad→Claude). Built, wired, and live-verified the narrow PreToolUse backstop personal/hooks/webfetch-github-guard.ps1 (denies WebFetch on GitHub source-file URLs only; generic research never matches).
  • settings.json portability. Uninstalled GitKraken's per-event Claude Code hooks (reversible; they were the main machine-specific churn); moved bell.ps1 into personal/hooks/; snapshotted the now-fully-portable settings.json to personal/claude-config/ (with a README); taught relink.ps1 to install-if-absent, flag GitKraken re-registration, and prune stale aggregation links; documented the new-machine reconcile (intelligent merge via the update-config skill) and the gk-disable step in new-machine-setup.
  • Design language + framing + consolidation. Captured dogfooding as first-class design language (CLAUDE.md design philosophy + the catalog) — Yggdrasil exists for Brad to dogfood AI, de-black-box it, and teach Claude to think the way that suits his autism. Raised the CLAUDE.md "learning AI" framing from beginner to "progressing." Generalized the learning-project bookmark (any subject; AI-learning as the first instance). Consolidated the brainstorming-skill review into the new "Next build" section.