dudgeon

Keep this project's graph lean and trustworthy — merge duplicates, retire stale notes (respecting task hierarchy), resolve contradictions, fix links, resolve entities, flag type drift, re-derive the per-entity timelines, regenerate the index-views, and surface stale tasks. Suggest-only; asks before deleting. Use periodically, before an important piece of work, or when the graph feels noisy. (brainkit's name for Karpathy's "lint".)

Capture new material into this project's graph as a typed entity — resolve what it mentions to canonical notes (with you, at capture), expand its shorthand and link it, preserve the raw, append to any touched entity's timeline, and extract the reusable knowledge and tasks. Use whenever the user shares or refers to something the brain should remember — a call, a meeting note, a doc, a decision, a person, a task, a link.

Put this project's graph to work toward the deliverable — answer a grounded question, advance a work/ deliverable with citations, run gap analysis ("what's missing for this?"), or read the task and reading-queue views — and file genuinely new conclusions back. Use when the user asks something the brain should know, asks to move a deliverable forward, or asks what's open or unprocessed.

Curate upstream improvements into this kit — a research task, not a script. Read canonical brainkit upstream, reason about what genuinely improved, and pull in only the changes you and the user find compelling, preserving local tweaks. Suggest-only; never bulk-overwrites. Use when the user wants to refresh the kit's machinery from origin.

Curate upstream improvements into this kit — a research task, not a script. Read canonical karpathy-core upstream, reason about what genuinely improved, and pull in only the changes you and the user find compelling, preserving local tweaks. Suggest-only; never bulk-overwrites. Use when the user wants to refresh the kit's machinery from origin.

Curate upstream improvements into this kit — a research task, not a script. Read canonical loopkit upstream, reason about what genuinely improved, and pull in only the changes you and the user find compelling, preserving local tweaks. Suggest-only; never bulk-overwrites. Use when the user wants to refresh the kit's machinery from origin.

Ingest a source into the Loop Library — read it, mirror it immutably, write its Source concept, and compile it into the wiki (loops/concepts/comparisons), updating indexes and the log. Use when the user adds a source (a paper, post, doc, transcript) or drops a file in inbox/ and wants it folded into the knowledge base.

Health-check the Loop Library wiki — run the conformance and lint scripts, then review for what scripts can't catch (contradictions vs newer sources, provenance drift, orphan/missing pages, stale timestamps) and fix what's safe. Use periodically, before merging, or when the user asks to clean up or audit the knowledge base.

Work with whatever is open in Duo — the browser pane, the markdown editor and the user's selection, the file navigator, the HTML canvas — and the user's work-notes vault (Duo's "graphbook" knowledge surface) — by driving the `duo` CLI. Use when the user references Duo's surfaces, asks to read or transform what's open or selected, open a local file, drive Google Docs / Sheets / Figma / Notion live, build an interactive page or playground, OR work with their knowledge vault / notes — capturing notes, filing or linking by type, querying the link graph (backlinks / orphans), or building rollup views / dashboards over note frontmatter — "roll up all my open tasks", "make a rich HTML rollup or dashboard of my initiatives", "list every note of type X", "what links here". A vault rollup/dashboard renders via `duo rollup render --html|--md` (works in OKF too) — NEVER a hand-built HTML page.

Generate an interactive HTML smoke-walk page for the user to validate a sprint's user-visible changes. Each item has pass/fail toggle + notes; the user clicks Copy, pastes the result here, and Claude parses it to flip statuses / propose a cut. PROACTIVELY OFFER at the end of any sprint that touched renderer/, electron/, cli/duo, skill/, or agents/ — before any cut-version proposal.

Generate a single-page interactive HTML worksheet for the user — items + per-item radio + textarea + sticky footer with Copy results / Send to Claude / Clear saved. Schema-driven; consumers (smoke-walk, sprint-plan, future retros / triage forms) just author a JSON manifest and feed it through generate.mjs. Use when you need structured, parseable feedback from the user that round-trips back into the conversation without you having to invent the UI each time.

Author, validate, and build a Duo distro pack — a Claude Code plugin folder with a Duo `duo-extras/` subtree that ships an organization's skills + agents + canvases + CLAUDE.md guidance to a canonical signed Duo install. Use when a user says "build a distro pack for X", "make an Acme-flavored Duo", "package my team's skills as a Duo plugin", or "scaffold a new distro pack." Pairs with the duo-pack-service install pipeline and the `duo pack list / uninstall` CLI verbs.

TODO — one sentence describing what this curriculum teaches and what the user walks away with after completing all modules. Concrete ("A 30-minute curriculum on Claude Code mental models, place, memory, and trust — for PMs new to agentic coding") beats abstract ("Onboarding curriculum").

TODO — one sentence describing what this lesson teaches. The user reads this in Duo's skill picker; keep it concrete ("Walks the user through Duo's three panes" — yes; "Onboarding lesson" — no).

Cut a new Duo version — drafts release notes (litmus test), bumps package.json if needed, rebuilds CLI, syncs skill, runs typecheck + smoke matrix, updates CHANGELOG, RELEASES.md, faq.html "What's New", what-duo-does.html new-capabilities, roadmap status flips, then commits + tags. PROACTIVELY OFFER when a stage flips to ✅, when a substantial commit lands on user-visible surfaces (renderer/, electron/, cli/duo, skill/, agents/), or when the user says "shipped"/"done"/"let's commit" on something user-facing. Geoff won't remember to ask.

Help the owner plan the next Duo sprint by harvesting candidates from tasks.md / active-sprint.md / session-log.md / roadmap.html, generating an interactive worksheet they prioritize (P0 / P1 / P2 / Skip), then synthesizing the result into a sprint commitment + active-sprint.md update. PROACTIVELY OFFER right after a cut lands or when the owner says "what's next?", "let's plan the next sprint", "let's triage", or signals end-of-chapter. Pairs with the `worksheet` skill (which it builds on) and `cut-version` (which it precedes).