misaon
UserA local multi-agent coding harness with a pixel-art office: AI agents work your repositories in isolated Docker sandboxes.
Categories
Indexed Skills (21)
agentic-engineering
Plan bounded work for the current floor and track concrete completion criteria.
context-budget
Reduce unnecessary instructions and tool output using measured session information.
search-first
Find existing repository code and maintained dependencies before creating new abstractions.
coding-standards
Review naming, boundaries, readability and maintainability in the current project.
security-review
Review concrete trust boundaries and reachable security failures in the assigned change.
verification-loop
Verify implemented behavior and report the checks actually performed.
bun-runtime
Use the project Bun version and verify runtime, bundler and dependency compatibility.
error-handling
Handle failures, cancellation and retries without hiding incomplete operations.
git-workflow
Commit work on the branch supplied by Home Office and preserve other work.
specify-work
Load at the start of every plan session, before the first ho_delegate. How to turn a request into tasks with observable acceptance criteria — reading the repository just enough, cutting along verification lines, wiring dependencies as data, choosing assignees by role, flagging QA and security review — and how to report the plan to the human.
delegate-work
Load before the first ho_delegate or ho_plan of a triage session, and whenever a request looks bigger than one task. How to route a request — an errand to the secretary, an obvious change to the developer whose role fits, everything else to the analyst — how to fill the ho_delegate fields when specifying the work yourself, and how to report.
review-session
Load at the start of every review session, before the skill of your role. The mechanics of reviewing a task's verified commit in a Home Office sandbox: reading the diff against the base branch in your own copy of the repository, collecting evidence per acceptance criterion, running what needs running, and filing exactly one ho_review verdict the author can act on.
work-session
Load at the start of every work session and again before ho_report. The order of work in a Home Office sandbox, from recovering context and setting up the repository through committing at checkpoints and running the floor's check the way the office re-runs it, to updating whatever documentation the change made untrue and a report written for the reviewers who read the branch next.
backend-change
Load in a work session when the task changes server-side code, an API, a database schema or stored data. How to make the change behind a stable interface, validate at the boundary, keep migrations safe and reversible, and prove the behaviour with tests before reporting.
general-change
Load in a work session when the task belongs to no specialist — a script, a build file, a CLI, a library, glue between two parts. How to bound the reading, take the convention from the code around you, change every call site rather than one, verify the way the repository verifies, and hand the task on when it turns out to belong to a specialist after all.
devops-change
Load in a work session when the task changes containers, CI pipelines, infrastructure definitions, deployment or developer tooling. How to make such changes pinned, least-privileged, reversible and verifiable from a clean clone.
frontend-change
Load in a work session when the task changes what a user sees or does in a browser or a graphical interface. How to build the change accessibly, keep it fast and visually stable, follow the repository's own components and tokens, and prove it in the browser before reporting.
review-branch
Load in a review session when the reviewer is the head of development, the last gate before the default branch. How to judge a task branch in order — design, functionality, complexity, tests, naming, consistency, documentation — building on the QA and security verdicts of the round, and how to write a verdict the author can act on.
test-branch
Load in a review session when the reviewer is the QA engineer. How to test a task branch against its acceptance criteria: run the checks and the application, exercise each criterion the way a user would, explore the edges and the unhappy paths, read the added tests, and write defects as reproducible bug reports.
small-errand
Load at the start of every work session of the secretary. How to do a small, well-defined errand — a changelog entry, documentation, a rename, a dependency bump, a mechanical edit — exactly as asked and with the smallest diff, and when to stop and hand it back instead of improvising.
browser-check
Load when the session has browser tools (browser_navigate, browser_snapshot) and the work or the review concerns something a user sees. How to start the app inside the sandbox, drive it with the Playwright tools, check the acceptance criteria and attach the evidence.
Bio shown is the top-scored skill's repo description as a fallback — real GitHub bios land in a future update.