Skills and session commands.

Everything that ships beside the qa CLI for your coding agent. This page is generated from the skills and commands themselves, so it cannot drift.

Where Jev is needed.

The qa CLI never calls Jev.

Static rules, browser probes, the Rust kernel and every exit code run on your machine. Some skills and commands add a screening pass by Jev, TypeSafe's System One model. Those need your own TYPESAFE_API_KEY (TYPESAFE_JEV_KEY is also read). The key is optional and user-owned.

When a Jev step runs, the state it screens is sent to api.typesafe.ai. Do not send code you may not share with that vendor. Every flag from Jev is a lead, never a finding, until code confirms it.

Skills

frontend-verify

skill
What it is for
Verify an application's frontend with static analysis, browser measurements, and explicit coverage gaps.
When to use it
For broken or stale data, cross-route consistency, pre-release QA, and installing the qa CLI. Use for bounded checks or explicitly time-budgeted, sustained QA; includes the former qa-research workflow.
Entry
Ask your agent to use the frontend-verify skill

Jev: not needed.

harvest

skill
What it is for
Turn a session's mistakes into executable guards — scripts and build-failing tests, not another doc nobody reads.
When to use it
At the end of a session, after a review round, or whenever the same class of bug has bitten twice.
Entry
Ask your agent to use the harvest skill

Jev: its screening steps need your optional TYPESAFE_API_KEY. State goes to api.typesafe.ai.

qa

skill
What it is for
One QA entry point, built on Jev (TypeSafe). DOCS consolidates docs with zero lost knowledge (claims checked against code, lost-claim review). HISTORY classifies a repo's past fix commits to show where bugs escape. HUNT screens every React/TanStack surface against generic checks and the repo's own invariants. REVIEW triages the open PR queue (stacks, PRs already contained in others, CI, risk) and assigns each PR a review depth and model (haiku/sonnet/opus) plus a merge order. PLAN writes an evidence-first QA plan from a diff. MEASURE runs the `qa` browser/static CLI.
When to use it
For "audit this repo", "bugs we keep missing", invariants, hooks/TanStack, edge cases, "review these PRs", "which model should do this", QA plans, pre-release QA.
Entry
Ask your agent to use the qa skill

Jev: its screening steps need your optional TYPESAFE_API_KEY. State goes to api.typesafe.ai.

ui-stress

skill
What it is for
Use when a UI must be tested against the states and content it will actually meet in production - empty, one, many, loading, error, permission-denied, slow, long text, broken image, 320px, dark mode - rather than the happy path it was built against. Triggers on "find the UI bugs", "test every state", "QA this screen", "why does it break with real data", "the empty state is broken", "it overflows on mobile", and on any codebase that has accumulated many UI defects and needs a systematic sweep instead of one-off fixes. Not for visual taste or design identity (use design-director), and not for building or redesigning UI (use impeccable).
Entry
Ask your agent to use the ui-stress skill

Jev: its screening steps need your optional TYPESAFE_API_KEY. State goes to api.typesafe.ai.

Session commands

/backend-edges

session command
What it is for
every endpoint judged by Jev, failing tests written for real gaps
Use it to
Backend edge-case hunt
Entry
/backend-edges repo path (default: cwd)

Jev: its screening steps need your optional TYPESAFE_API_KEY. State goes to api.typesafe.ai.

/checkpoint

session command
What it is for
commit, push, checkpoint log with verify certificates, status. For leaving, risk, or a session that feels wobbly
Use it to
Force a save-point NOW
Entry
/checkpoint

Jev: not needed.

/featuredev

session command
What it is for
enumerate every feature, test, fix, retest, with an honest done-checklist
Use it to
Feature-inventory loop
Entry
/featuredev

Jev: not needed.

/map

session command
What it is for
conclusions to docs, state never
Use it to
Explore a repo in parallel and persist ONLY the stable map
Entry
/map

Jev: not needed.

/mission

session command
What it is for
phases with per-phase wraps, batched blockers, self-preservation before limits, skeptical review before final wrap
Use it to
Long-running autonomous run
Entry
/mission

Jev: its screening steps need your optional TYPESAFE_API_KEY. State goes to api.typesafe.ai.

/review-queue

session command
What it is for
Jev picks depth and model, code picks merge order
Use it to
Triage and review every open PR in a repo
Entry
/review-queue OWNER/REPO (default: this repo's origin)

Jev: its screening steps need your optional TYPESAFE_API_KEY. State goes to api.typesafe.ai.

/start

session command
What it is for
preflight (git/gh/PATH), enumerate sources with counts, set the working rules (commit/push/checkpoint continuously, verify at source, volunteered status)
Use it to
Open a session
Entry
/start

Jev: not needed.

/ui-hunt

session command
What it is for
history → shared causes → browser proof
Use it to
Find the UI bugs this repo keeps shipping
Entry
/ui-hunt repo path (default: cwd) [--base URL]

Jev: its screening steps need your optional TYPESAFE_API_KEY. State goes to api.typesafe.ai.

/wrap

session command
What it is for
land all work, derive changelog/issues/board from git and gh, hand off into the PR, final gate-output status (no percentages)
Use it to
Close a session
Entry
/wrap

Jev: its screening steps need your optional TYPESAFE_API_KEY. State goes to api.typesafe.ai.