eugenelim

Use this skill whenever you're implementing a non-trivial change -- a feature, a multi-file bug fix, a refactor, a migration, a framework or dependency upgrade, a schema or API change, performance work, an infrastructure or build-system edit, or anything spec-driven. Also triggered by argless resume phrases -- "resume", "continue", "keep going", "pick up where I left off", "let's get going" (bare phrases only; "resume the X project" or "resume the X investigation" routes to desk-research-project-status). It enforces the project's plan -> execute -> self-review -> fix loop with mechanical gates (lint, typecheck, tests) and adversarial review. Default to this skill for any task larger than a one-line edit.

Use when holding a raw signal (market observation, OKR gap, user pain pattern, engineering finding, competitive signal) at initiative or capability level — turning it into a typed finding, a Wardley capability maturity assessment, and an anchor into the PE six-step shaping sequence. Triggers on "frame this situation", "what do we do with this signal", "is this an opportunity or a risk", "where should we start", "map this to the shaping sequence". Do NOT use for a feature-scoped request (use `frame-intent`), testing an assumption (use `de-risk-intent`), or breaking down a shaped intent (use `decompose-intent`).

Use this skill when a session has surfaced a list of future work — follow-ons, review recommendations, audit remediation items, deferred scope — and you want to capture it into workspace.toml so a later session can pick it up cold. Triggers on "capture this", "add these to the queue", "capture these as queue items", "queue these up", "add this to the backlog" + a bulleted or numbered list in context. Do NOT use to turn unstructured external input into a product brief (use author-brief), to decompose a brief into specs (use receive-brief), or to orient at session start (use workspace-status).

Use this skill to orient at session start, check initiative queue state, or see what's ready to work on next. Reads workspace.toml and surfaces ready-to-start items, blocked items with reason, parallel candidates, and active signals. Triggers on "workspace status", "where am I", "orient me", "session start", "what's ready", "show the queue", "what's next", "what should I work on", "check workspace", or any cold-start orientation request. Offers to initialise workspace.toml if absent.

Use this skill when the user receives an externally-authored, multi-feature product brief -- a PRD, a solution handoff, a requirements packet -- and needs to turn it into shippable specs. Triggers on "receive a brief", "decompose this PRD", "we got a product brief", "break this handoff into specs". Elicits the load-bearing fields without mandating a schema, decomposes by shippability, and executes each slice through new-spec then work-loop. Do NOT use to author a single feature from scratch (use new-spec) or to record a decision (use new-adr).

Convert documents and images to Markdown for AI context layers. Documents (PDF, DOCX, XLSX, PPTX, HTML, EPUB, CSV/TSV, OpenDocument, .eml) go through `scripts/convert.py`, which extracts at a no-ML Tier-0 floor (pure-Python / stdlib parsers) and falls through to Docling (Tier 2) only for .xls and images; images go through a two-pass sliding-window vision pipeline whose tiling and reconciliation are deterministic (`scripts/split_image.py` and `scripts/reconcile.py`). Every output carries a versioned frontmatter contract (provenance + a quality/confidence signal). The agent's job is the per-tile vision read; tile dedup and ordering are handled by the script.

Use this skill when the user wants to start a new feature with a spec, or wants to write a spec for something they're about to build. Triggers on "new spec", "write a spec for X", "let's spec this out", "start a feature for...". Spec-driven development; the spec drives implementation. Do NOT use for cross-cutting proposals (use `new-rfc`) or recording decisions (use `new-adr`).

Load when a task's primary output is HTML, CSS, or JS. Provides design pre-flight, codified craft rules, and GATES verification commands for that surface.

Unified entry point for ingesting any source (paste, URL, file, photo, transcript) into the vault. Routes on two axes — source-type (clean the input via the document/website/inline path) and content-type (apply schema and routing via a content-type primitive's ingester). Use whenever a user drops, pastes, or links any source for ingestion.

Run health checks on the vault. Combines structural checks (broken wikilinks, orphan pages, missing frontmatter fields, stale `modified` dates, tag synonyms, convergence debt in `raw/`, asset coverage) with semantic checks (contradiction detection across pages claiming the same fact). Use on request — "lint the wiki" / "check wiki health" — weekly or per-sprint, or after bulk ingestion of new raw sources.

Search the vault by content and frontmatter (--genre, --subtype, --tag, --status, --top). Returns ranked pages with title, genre, subtype, status, tags, and per-file match count — ready to read. Today's backend is ripgrep (literal substring, zero install); the SKILL was designed against a future SQLite FTS5 auto-upgrade tier (BM25, stemming, phrase/prefix) that has not yet shipped. Use for any *vault content* question; reserve the IDE's built-in Grep for regex, code search, or inspecting a known path.

Scan the vault for follow-up callouts (`> [!important] Follow-up due by YYYY-MM-DD`) and standalone `action-item` pages, then surface what's overdue and what's due in the next 60 days, grouped by owner and domain. Load when the user says "what follow-ups are due?", on a weekly sweep, or via `wiki run follow-up-tracker`. Writes one report page per run; pairs with `medical-summary` (which produces visit-specific summaries) and the content-type ingesters that emit the callouts in the first place.

Ingest a decision source (meeting decision callout, email thread that landed on a choice, written rationale or short ADR-style doc) into a structured decision page. Load from the `ingest` skill when content-type routing identifies the source as a durable choice with context, alternatives, and consequences — distinct from a meeting (which captures *what happened*) or a stakeholder-update (which captures *what to communicate*). Produces one page under `wiki/library/`, links the owner to `wiki/people/`, and is read by downstream operations (status-synthesis, onboarding-pack).

Ingest a meeting source (transcript, notes paste, recording link) into a structured meeting page. Load from the `ingest` skill when content-type routing identifies the source as a meeting (calendar invite text, transcript turns, agenda+notes shape). Produces one page under `wiki/library/`, links attendees to `wiki/people/`, and registers decisions and follow-ups for downstream operations (weekly-digest, follow-up-tracker).

Ingest a trip booking confirmation (hotel, flight, rental car, activity reservation) into a structured `trip-doc` page in the trip's container folder `efforts/trips/<slug>/`. Load from the `ingest` skill when content-type routing identifies the source as a booking (confirmation email, PDF, or web page). Aggregates: a single trip typically gets several runs — each new booking appends to the same trip page. Pairs with the `trip-prep` operation (which writes packing + pre-departure tasks back to the same page 2-3 weeks before departure).

Produce a weekly family meal plan with shopping list, reading the recipe pages in `library/` (`--subtype recipe`), the dietary-notes reference page, last week's plan, and family calendar context if available. Load when the user asks for a weekly plan or to plan-meals, when `wiki run meal-planning` invokes you, when `/plan-meals` fires from Claude Code's slash palette, or on a scheduled Sunday sweep. Writes one page to `outputs/meal-plans/<window>.md`; idempotent for a given window. For ad-hoc "what's for dinner tonight?" use a recipe recommendation skill instead — this operation is the weekly cadence.

Use this skill to draft an email reply (or reply-all / forward) in the user's voice for their review, saved to the mailbox Drafts folder via the inkwell CLI. Triggers on "draft a reply", "respond to this email", "write a response", "reply to <sender>", "answer my reply-later stack", "forward this to Bob with context", "set my out-of-office". Produces a draft for approval with a confidence label — it never sends (Mail.Send is denied by construction, PRD §3.1; the user sends from Outlook). For triaging use `mail-triage`; for read-only briefings use `mail-executive-summary`.

Use this skill to produce a morning email briefing for a busy senior leader from their actual mailbox via the inkwell CLI — what needs their reply, what they're waiting on, what's worth knowing, and what's already handled, plus calendar. Triggers on "brief me on my email", "what needs my attention", "morning inbox rundown", "catch me up on my inbox", "summarize my unread", "what's on my calendar today". Strictly read-only — never archives, deletes, routes, drafts, or marks read. For acting on mail use `mail-triage`; for writing replies use `mail-draft-reply`.

Use this skill to find threads where the user is waiting on someone else and hasn't heard back, then draft polite nudges to chase a response — via the inkwell CLI. Triggers on "follow up on", "chase", "nudge", "what am I waiting on", "remind people who owe me a reply". Drafts for approval, saved to the Drafts folder — it never sends (Mail.Send is denied by construction, PRD §3.1; the user sends from Outlook). The action counterpart to `mail-executive-summary`'s "waiting on others" section.

Use this skill to prepare a concise briefing on the external people the user is about to meet — who they are, their company, recent discussions, and anything outstanding — via the inkwell CLI (Graph calendar view + cached mail), optionally with web research. Triggers on "prep me for my meetings", "who am I meeting today", "brief me before this call", "tell me about <person> before our meeting". Read-only: never sends, replies, or modifies the calendar. For inbox briefings use `mail-executive-summary`.

Use this skill to free up mailbox storage via the inkwell CLI — soft-delete (move to Deleted Items) the junk that takes the most space, large attachments first, then old bulk-sender back-catalogs and stale low-value mail. Triggers on "free up storage", "my mailbox is full", "clear space", "trash junk to reduce storage". It never permanently deletes and never empties Deleted Items — the user does that. Confirm every batch. For volume (not size) cleanup use `mail-triage`; to stop the junk arriving use `mail-reduce-noise`.

Use this skill to cut recurring inbox clutter at the source via the inkwell CLI — find high-volume, rarely-read senders and apply a standing decision (unsubscribe, auto-route past the inbox, or keep) so their future mail never piles up. Triggers on "reduce email noise", "unsubscribe from junk", "clean up my subscriptions", "stop newsletters cluttering my inbox", "which senders are noisiest". Different from `mail-triage` (one-off processing of what's already there) — this changes what arrives. For a one-off cleanup use `mail-triage`.

Use this skill when the user wants their actual mailbox triaged, organized, or processed via the inkwell CLI — acting as an executive assistant on live mail, not developing inkwell. Triggers on "triage", "sort my inbox", "organize my email", "process my inbox", "clear my inbox", "clean up my email", "deal with the screener queue". Makes one classification pass and files every inbox thread into the workflow folders (@Action / @Waiting / Archive / Reference / Newsletters) or soft-deletes the worthless — behind a presented list and explicit confirmation. Do NOT use for inkwell code changes (use `work-loop` / `bug-fix`), for read-only briefings (use `mail-executive-summary`), or for writing replies (use `mail-draft-reply`).

Use this skill to walk the adopter through the four classes of post-install change (substitution, .upstream companion merges, discovery + restructuring, within-layout consolidation). Triggers after installing a pack (the install->adapt chain nudges via session-start hook) or any time `<repo>/.adapt-install-marker.toml` / `~/.agentbundle/.adapt-install-marker.toml` is on disk. Walks both scopes' state files for Tier-2 detection; class-1 substitution shells out to `agentbundle adapt`; classes 2-4 write files directly under the per-scope path-jail.

Use this skill when the user asks to create, write, draft, or open a new ADR (architecture decision record). Triggers on phrases like "new ADR", "write an ADR for…", "record this decision", "let's ADR this". Do NOT use for RFCs (use `new-rfc`) or for the phased build plan (that's docs/PLAN.md).

Use this skill when the user wants to propose, draft, or open an RFC (request for comments). Triggers on "RFC", "propose a change to…", "let's get input on…", "draft a proposal". Do NOT use for already-decided things (use `new-adr`) or for changes that fit in a normal PR.

Use this skill when the user wants to change `docs/CHARTER.md` or make significant changes to `CONTRIBUTING.md`. Triggers on "let's change the charter", "update the principles", "amend the charter", "change how we work". Charter and significant convention changes go through RFC review, not direct PR.

Use this skill whenever you're implementing a non-trivial change — a phase of the build plan, a bug fix touching more than one file, a refactor, or anything driven by a conformance case. Enforces the plan → execute → gates → review loop with mechanical verification. Default to this skill for any task larger than a one-line edit.