wayfare

Advance one item as far as its gates allow: a ready task through the build pipeline, a Dependabot PR to merged, or one turn of a goal. Never plans. Use to move a specific item forward by id.

Run the compliance register audit on its own (this repo against the baseline, or the whole fleet at merged state), report register defects, regenerate CONSISTENCY.md, and draft backports to the template. Use to check conformance without a full roadmap sync.

Audit the wayfare plugin. Checks skill quality, consistency, DRY violations, HERO.md field coverage, and readability. Use before releasing changes to the plugin.

Run by wayfare-sync-plan. Audits the codebase read-only for hardening: dependency CVEs, container CVEs (Scout and Trivy), and code-level robustness. Emits execution-ready plans as .plans security items and never edits source.

Drive a task end to end: plan, implement, simplify, push (tests included), self-review, mark ready, await review, respond, ship. No args: resume the current goal (gated). Use for small, low-risk PRs only; larger work goes through wayfare:wayfare-sync-plan.

Run pre-flight checks for the wayfare pipeline. Catches missing tooling, stale HERO.md, .env mismatches and busy ports before any step does destructive work. Use before wayfare-push-pr, wayfare-ship-pr or wayfare-build-task, or when a pipeline step fails on setup.

Create a new Claude Code skill, subagent, rule, or hook. Guides through trigger conditions, success criteria, and writes a well-structured SKILL.md. Use when extending Claude.

Abandon work on an unmerged branch and write status dropped on the item, so the roadmap stops claiming it. Stashes only with a named, confirmed stash. Use when work on a branch is being given up rather than finished.

Brainstorm and grill an idea one question at a time into shared understanding, as dependency-aware work-items. Use when starting a feature, refactor or migration past a one-liner, when a task arrives vague, before an expensive-to-reverse decision, or before building on an unstated assumption. Skip it for typos, copy tweaks and dependency bumps.

Remove signs of AI-generated writing from text so it reads as human-written; based on Wikipedia's "Signs of AI writing". Use when editing or reviewing prose, or on any text you want to pass through the same filter the pipeline applies.

Configure a repo for wayfare: investigate it, write HERO.md, and create the plan object .plans/PLAN.md, migrating an older store on sight. Scaffolds first in an empty directory. Use on a repo that has no HERO.md or no .plans/PLAN.md.

Test (verify + smoke), commit, push, and open a draft PR with a CI report. Use when work is ready to commit or share. Pass test for the test phase only, commit to stop before pushing, ready for a non-draft PR, or a target branch to merge into.

Report and tune the config every wayfare skill reads: the `## Connections` blocks and `## Wayfare` in HERO.md plus the fields the architecture and security stages read. Asks only about fields that are unset or wrong, then writes. Use when a run complained about config, or after the repo changed shape.

Refactor a project's UI into atomic components (atoms/molecules/organisms/templates), sourcing primitives from a design-system registry when one is configured (default @aihero) or stock shadcn otherwise, and codemod off-token styling. Use when asked to recomponentize, refactor UI, adopt a design system, or clean up component structure.

Read PR review comments, fix the code issues they raise, and resolve the conversations on GitHub. Handles the full respond-to-feedback cycle. Use when a reviewer or review bot has left comments on your PR, or when asked to address feedback.

Report where DESIGN.md and the codebase have drifted apart: claims the code no longer backs, layers the file never mentions, sections describing something that was dropped, and a missing or stale Source ref. Writes nothing. Use to check the architecture record is still true before trusting it.

Report drift between FLEET.md and the checkouts beside it, read-only. Names repos missing from the map, rows with no checkout, and port collisions between dev stacks. Writes nothing. Use from the folder that holds the repos, to check the map before trusting a fan-out.

Review a PR with the pr-review-toolkit agents plus a security pass. No argument reviews your own draft PR and applies fixes. A PR number reviews the author's code via inline comments, no edits. Use before marking your PR ready, or to review someone else's.

Set up a developer's local environment. Reads HERO.md, checks required tools, guides through git config, CLI auth, and missing dependencies. Per-developer, and never modifies shared files. Use on a fresh machine or clone, or when a tool or login is missing.

Trigger auto-approve on a PR, wait for the verdict, merge if it passes, and reset to the default branch. Use after wayfare:wayfare-review-pr and any bot review when the PR is ready to ship.

Authorize the next goal at a gate the person types, then run its first turn. The grant is in-session and never stored. Use when the roadmap is planned and you are ready to start building the next goal.

Converge DESIGN.md with the codebase: bootstrap it where it does not exist, and apply the drift rows a review found, writing only what the user confirms. Decisions are append-only. Use after wayfare-review-architecture reports rows, or to create the record for the first time.

Create and converge FLEET.md, the local unversioned map of sibling checkouts (group, port). Scans the folder, proposes rows, and writes only what the user confirms. Use from the folder that holds the repos, when adding a checkout or claiming a port. To report drift without writing, use wayfare-review-fleet.

Converge the roadmap with the world: the architecture record, the design snapshot, the hardening audit, prose gone false about the code, the compliance register, the dependency bots' PRs, branches built and never merged, and goals, into .plans. Proposes items and goals; writes only what the user confirms. Use to refresh what is worth doing.

Distill the current conversation into one self-contained work-item covering context, decisions, remaining work and acceptance criteria, for a downstream agent with zero context from this session. Use when stopping mid-task, handing work to someone else, or filing what was learned as a ticket.