Hone
Close the grade→improvement loop — read accumulated grade-session report cards, find failures that RECUR across sessions, and propose surgical edits to the skill bodies that caused them; write only on approval. The suite's self-improvement layer (grade-session diagnoses per session; hone sharpens across sessions). Modes: sweep | audit | <skill-name>.
Pmf
Alias for /ground — the demand-evidence gate on the build decision. Binds every product bet to real demand evidence (revealed-&-costly > specific-stated > vanity proxy) and REFUSES to license a build on anything weaker than specific stated demand: defaults to UNKNOWN/PUSH and ships the cheapest test that would settle it. Stops craft drifting into impressive-but-unwanted. Modes: <bet|product> (ext…
Post
Posting strategist for the build-in-public pool — surfaces the next post (inventory- and sequence-aware), produces register-true copy per platform in the user's voice, and hydrates the week's Google Calendar with ready-to-paste events (full copy + media paths). Not a generic repurposer. Modes: next | week [N] | log <url>.
Stash
Zero-friction capture inbox for the thought worth keeping that flies past mid-task — park it in one line, keep working, drain it later into the gated skill that owns its destination. The missing low-friction front door to the flywheel (capture is trivial; the DRAIN is the product). Modes: <note> (capture) | drain | list.
Ground
Demand-evidence gate on the build decision — bind every product bet to real demand evidence so craft does not drift into impressive-but-unwanted. Classifies evidence against a hard hierarchy (revealed-&-costly > specific-stated > vanity proxy) and REFUSES to license a build on anything weaker than specific stated demand: defaults to UNKNOWN/PUSH and ships the cheapest test that would settle it. T…
Architecture Blueprint
>-
Taste
Distill the user's revealed taste from real session transcripts into an enforceable personal rubric (~/.claude/TASTE.md), then apply it as a pre-delivery filter. Mines what they actually corrected and accepted — never what they say they prefer. Modes: induce | apply | audit.
Until
Goal-pursuit engine — attempt, verify, adjust until a pinned objective is verifiably done. Negative-control verification, a hypothesis ledger that forbids re-testing refuted ideas, git checkpoints, and a forced escalation ladder (switch layers → fresh-context subagent → attack the goal itself). Survives session death via a state file. (For time-based recurrence, use the built-in /loop instead.)
Harmonize
Internal-consistency enforcer — propagate a NAMED standard to every part. Where /frontier sets each surface's standard from the outside world (outward), harmonize sweeps INWARD for parts that drifted from our own declared standard (a DESIGN_SYSTEM rule, a CLAUDE.md mandate, the house skill template, an output contract) and converges them. Requires a named reference — that constraint is what keeps…
Snip
Save a code/md snippet as a build-in-public bundle — source + freeze PNG + draft post caption — under ~/build-in-public/snippets/, with a disclosure-boundary and secrets check first.
Raze
Apply the razor — raze a target to the simplest sufficient form by QUESTIONING every requirement and DELETING whole parts/processes before any optimization (Musk's Algorithm: question → delete → simplify → accelerate → automate; idiot index; the best part is no part). A "should this EXIST at all" scythe, not a tidy-up. Deletion is the judgment act. Modes: <target> (scan+propose) | audit | apply.…
Tidy
The judgment layer for workspace hygiene — resolves the loose ends the hygiene hook SURFACES with the reasoning a bash reconciler cannot do: branch-supersession triage (merged-equivalent-via-squash vs real-unmerged vs stale-divergent vs sibling-superseded), push/PR decisions, worktree-removal calls, and PR-state awareness. The on-demand judgment sibling of the always-on workspace-reconcile hook (…
Harvest
Harvest the session that's ending — extract every durable asset (transcript + findings + decisions + a candid graded report card + build-log + memory) so the next run starts richer. Renamed from /archive-session; folds in /grade-session's grade.
Page Audit
Audit a local HTML page's layout and rendering the trustworthy way. Measured facts (overflow, element widths) via an instrumented iframe harness, then screenshots via modern headless Chrome with animations neutralized. Use before diagnosing any visual bug, and instead of eyeballing headless screenshots.
Redline
Convert a NEAR-MISS into an ENFORCED guardrail — a mechanism a non-agent actor (the harness/hook runtime, git, CI, the test runner, the OS) executes automatically and that BLOCKS on violation, not a passive lesson the agent must remember. The active-enforcement layer: where /harvest records a lesson and /hone advises from recurring grades, redline makes the dangerous class structurally impossible…
Card
Turn a finding, insight, or stat into a designed, share-ready PNG card — modern product-design aesthetic (house visual system in ~/.claude/DESIGN_SYSTEM.md), one idea per card. Renders HTML/CSS via headless Chrome. Complements /snip (code windows): /snip shows source, /card stages an idea.
Capitalize
Put accumulated capital to work at session start — withdraw the deposits harvest made (findings, decisions, open questions, until-ledgers, lessons) and deploy the slice relevant to THIS task as a ready-to-execute brief, so a fresh session starts warm instead of paying rent to re-learn the project. Then compound it: promote dormant assets into standing structure. Reads the archive, writes nothing…
Milk
Tap ONE rich asset for durable yield — point it at a deep-research report, a competitor repo, a paper/PDF, a long thread, or a codebase you're holding RIGHT NOW, and squeeze every durable drop (findings, decisions, saveable prompts, postable snippets, lessons, open questions) WITHOUT ending the session. Routes the yield into your existing capital stores so the squeeze feeds the flywheel. Re-milka…
Moonwalk
Walk an artifact BACKWARD through time to recover, with evidence, WHY it exists — provenance archaeology + a Chesterton's-Fence verdict (KNOWN-DEAD / KNOWN-LIVE / UNKNOWN) that gates a delete or a guarded edit. The evidence-gatherer /raze's fence-gate needs and lacks: it makes "tests pass, cut it" survivable by surfacing the untested failure mode a guard encodes. Reconstructs the cross-store link…
Commit Mine
Commit ONLY this session's work out of a shared dirty tree where parallel agent sessions have uncommitted edits in the same files. Positive hunk selection, staged-snapshot test (checkout-index + suite), and a foreign-symbol check before every commit. Use whenever git status shows modifications you didn't all make.
Calm Ui Audit
>-
Nightshift
The night shift — call it at bedtime. It builds its own worklist across your active projects (stalled pursuits, loose ends, dead ends, refactors, research runs, efficiency audits, design-system fixes), executes overnight under an absolute safety contract (branches only, nothing irreversible, nothing outward-facing), verifies and adversarially reviews everything, and leaves a 2-minute morning brie…
Burn
Token-economics engine — audit where a session's tokens actually went (with $ math from real transcript usage data), pay down "session rent," and enforce structural spend disciplines that cut cost without touching quality. Built on agentic billing mechanics, not "be concise" tips. Modes: audit | rent | charter.
Prompt
Personal prompt library + prompt compiler. Save frequently used prompts as parameterized templates, invoke them by name, or refine a draft prompt into a more effective, token-lean version — informed by TASTE.md so the refined prompt asks for what the user will actually accept, not just what they typed. Modes: <name> | save | list | refine.
Grade Session
Grade the session that's ending — two candid report cards (the user's prompting, the agent's performance) plus a coach note for each side. Prints in chat; optionally saves to a file.
Mirror
Sync a skill from ~/.claude/commands/<x>.md to the public skills repo (yiyaw-lab/agent-armor; local clone ~/code/claude-code-skills) as <x>/SKILL.md — disclosure+secrets gate first, README bookkeeping (badge count, playbook diagram, catalog row) handled, then a PR with CI green; merge only on explicit approval. Curated: only named skills, never a bulk export. Modes: <skill-name> | sweep | audit.
Publish
Schema-first publisher for yiya.dev. Turn one durable, public-safe artifact from a Claude Code session into a typed content entry, after disclosure checks, a secrets scan, and the four-part quality gate. Reads live yiyaw-lab/yiyadev schemas, proves the entry with npm run build, then opens a PR or compare URL. Never merges. Modes: <artifact|path> | sweep | draft.
Continuity Context
>-
Relay
Credit-aware long-run carrier — take a multi-step autonomous chain across a credit/rate-limit window, a planned pause, or a budget ceiling WITHOUT losing work. Checkpoints durable state (commit+push branches with verified pushes, persist the todo list + worktree paths + a resume brief), schedules a session-death-surviving wake, auto-resumes single-flight on the other side, and finishes the checkl…
Elonize
Alias for /raze — apply Elon's principles (Musk's Algorithm: question requirements → delete → simplify → accelerate → automate; idiot index; the best part is no part). The razor (delete to the simplest sufficient form) is the core move; this is the growing basket of those first-principles rules. Modes: <target> (scan+propose) | audit | apply.
Vacation
Extended UNATTENDED autonomous build mode for when you're away (the multi-day sibling of /nightshift). Auto-runs on a schedule, builds features from the PRD/backlog on branches, and OPENS PRs for your async review — never merging. It earns more autonomy than nightshift via an async escape valve: when a decision is genuinely ambiguous it TEXTS you through your always-on agent, parks that task, and…
Full Throttle
Turn credits that would EXPIRE UNUSED at a reset into useful, banked work — the burn-it-forward sibling of /relay. When a session/credit window is about to reset with budget left over, fan out the maximum USEFUL, disk-banked, credit-intensive work as background jobs that survive the boundary, then checkpoint what's running and how to aggregate it after. Honest by construction — never fakes a cred…
Frontier
Bring your work to the frontier — across any SURFACE (design, model/LLM stack, agent architecture, infra, content...). For a surface it audits where you have fallen behind the live 2026 standard (benchmarked against that surface's frontier exemplars) and, on approval, provisions the current frontier toolchain. The external standard-SETTER; harmonize is the internal-consistency enforcer it pipelin…
Vault
Prompt vault plus pre-enrichment policy. Save high-value reusable prompts, preserve the original, write a stronger reusable version, add trigger metadata for future retrieval, and decide when broad prompts should be pre-enriched before implementation. Modes: save | enrich | fire | list.