Operating System

The tools and methods have been evolving faster than they've been consolidated. A three-way audit (philosophy docs, tooling inventory, cycle-execution history) found the composition-level philosophy alive in the work but the process side violating the methodology's own stage rule ("early stages need more freedom, later stages need more order"): archiving dead since c50 (June), 30 "active" cycles with ~7 in motion, a 55-file uncommitted surface, stale/conflicting commands, and no mechanism that actually imparts the methodology to agents — the "instructions vs. literacy" distinction from brand/design_philosophy.md is unrealized. The name comes from the philosophy's own line: "the design methodology functions as an operating system for human-AI collaboration."

Supersedes and absorbs c53 (agentic-workflows), which captured this same audit in July and was parked until Mark could give it full attention.

Three tracks, all in scope:

1. Order restoration — archive sweep drains CYCLES.md to genuinely-active cycles, HISTORY.md revived and cleaned, backlog committed, PICKUP.md + thread conventions codified in cycle-manager and synced to sister projects.
2. Tooling repair — commands converted to proper skills with frontmatter, figma-to-vue collision resolved, stale content fixed (release-layer, document-pattern, tour, archive), /pickup and /retro skills added.
3. Literacy layer — condensed methodology section in CLAUDE.md so every session carries it, pointers from building skills, /retro auditing adherence periodically.

Done looks like: CYCLES.md lists only in-motion/parked cycles, git status is clean, a fresh session knows the methodology, and the skill roster is accurate and self-consistent.

• cycles/CYCLES.md, cycles/HISTORY.md, per-cycle CYCLE.md Outcomes
• .claude/agents/cycle-manager.md (canonical) + the 6 sister copies
• .claude/commands/*.md → .claude/skills/<name>/SKILL.md (+ new: pickup, retro)
• CLAUDE.md (methodology section, Tour Guide Sync section)

Out: sister-project literacy propagation beyond the cycle-manager/tour sync; executing c53's automation workstreams (hooks, scheduled automation, plugin distribution) — those remain future cycles.

• 2026-08-15 — Cycle created from Mark's retro prompt; c53 absorbed and closed. Full audit findings in evaluation.md; execution plan in operating-system-plan.md.
• 2026-08-15 — Order restored. Backlog committed in five coherent commits (c62 visibility stream, c67 teams, c68 index, c66 cost, registry catch-up). Archive sweep: 22 cycles closed into HISTORY.md with Outcomes written where missing (c17, c33, c37, c40, c48, c54, c60) and c63's CYCLE.md backfilled; HISTORY.md duplicate c25/c12 entries removed; CYCLES.md drained 31 → 16 entries with status parentheticals. Six cycles flagged for Mark's call: c41, c42, c44, c51, c55, c59. c49 kept (blocked on Cloudflare secrets).
• 2026-08-15 — Conventions codified. cycle-manager gained Park/Resume ops with the PICKUP.md spec (generalized from c62), the real CYCLES.md entry format, and an optional Mode: template line; canonical synced to all 6 sister projects (extensions preserved) and committed in each repo.
• 2026-08-15 — Tooling modernized. All 9 commands converted to skills/<name>/SKILL.md with frontmatter (market-analyst pattern), commands/ removed. figma collision resolved: command → figma-to-component (layer components), skill stays figma-to-vue (page mocks). release-layer rewritten to verified reality (GitHub Packages, tag-driven publish.yml, update-consumers.sh — tarball flow fully retired); document-pattern/scan-patterns updated to the pattern-registry structure; archive's MCP rebuild step verified REAL (mcp/src/resources.ts imports archive-meta) and kept with justification. New skills: /pickup, /retro.
• 2026-08-15 — Literacy layer. CLAUDE.md gained a Design Methodology section (wicked problems, assemblage, modularity, edge of chaos + stage rule, 7 modes, 5-question framework, replicate-first, demo-is-the-test) and the missing Tour Guide Sync section; building skills carry a one-line methodology pointer.
• 2026-08-15 — Content-shots thread (content-shots-prompt.md / -plan.md). Capture script gained content mode (default): hides nav/footer + the page's title/preamble block via a structural rule ("Source:" paragraph anchors the preamble; fallback: first child + trailing <p>s), so heroes open on the visual work; --chrome keeps site chrome; --with-full saves a -full.png when content runs past the 1100px fold. Sweep + curated set recaptured in place (same filenames — no history-page edits); 17 full-page variants; archive entries gained two-image galleries (hero + full; detail view shows gallery instead of image, thumbnails keep the hero). Manifest now carries contentHeight/hasFull for the cycle viewer. Total 2026-q3 imagery: 18 MB. (history-refresh-prompt.md / -plan.md). Built scripts/capture.mts (npm run shots — Playwright against the running dev server, page-registry as manifest, ThemeEditor/devtools hidden): 90/90-route sweep + manifest.json into public/archive/2026-q3/sweep/ (11 MB), 20 curated cycle-named shots. Archive revived with 20 content/archive/2026-q3/ screengrab entries (cycle tags, feeds /timeline). All six /history pages backfilled — design +18 entries (c39→c69, 44 total), qmap +4, qflow +3, bob-wire +4, market-analyst +3, qpixel +1 (sister pages via parallel subagents from their own cycle records) — and hub cards corrected. Cycle-viewer architecture prepped: header parsing contract, cycle-prefixed filenames, manifest, archive cycle tags. Deferred: cycle? field on page-registry (file carries another session's uncommitted work); the reserved 'history' timeline source kind (c39 backlog).