JosiahAllard
@JosiahAllard
Public Skills
backlog-ops
by JosiahAllard
Use when you need to read or change a repo's backlog store with <kit>/tools/backlog.py - next, show, add, pull, close, block, set (including narrowing a record's scope). Gives the forms that pass the safety checks and what to do when one is refused.
land-branch
by JosiahAllard
Use when you are a lander told by the coordinator to land one reviewed branch on the main branch, and as a coordinator before queueing a branch for landing and after every landing. Covers <kit>/tools/land.py --submit, the wait-result verdicts, read-only verification, separate cleanup, what to do on a conflict, a red gate, a stuck landing or a safety-check refusal, and the coordinator's queueing, record closing and after-landing notices.
session-handoff
by JosiahAllard
Use when the PM tells you to "checkpoint and stand down" for an account hand-off (coordinator or long-running session), when a coordinator winds down a finished work set, and when you are the successor session started after one. Covers writing the state note, the exact ready reply, and the successor's verify-then-act start.
godot-render-check
by JosiahAllard
Use in a Godot repo when you must render a frame to look at it (a preview scene) or run the end-to-end tier. Covers light vs heavy, ports, the machine lock, orphan sweeps and the pass/fail strings. Repo values (preview scenes, ports, scripts, strings) come from the repo's own docs, not from this text.
branch-sync
by JosiahAllard
Use when the coordinator says the main branch moved and your branch must catch up. Covers merging the exact named commit, what never to do (rebase, local main, force-push), conflicts on your own branch, and re-gating.
branch-review
by JosiahAllard
Use as a coordinator before giving any branch its landing verdict. Covers the scope check, when to read the diff yourself and when to dispatch a reviewer, what to look for, sending findings back, the three verdicts, follow-up records, preventing a defect class seen twice, and CI changes that only add strictness.
machine-lock-run
by JosiahAllard
Use before any heavy work on the shared machine - builds beyond the ordinary gate, mutation or stress runs, LAN or lossy e2e, renders, local-AI generation, training - to wrap it in the machine lock or its background mode, read status, and handle busy and reservation refusals.
doc-claims-discipline
by JosiahAllard
Use when you record a finding or conclusion, withdraw one, trim or delete lines from a doc, or add a capability that earlier docs may say does not exist. Covers evidence-labelling, reversals living in the file, what a deletion silently removes, and stale absence claims.
coordinator-dispatch
by JosiahAllard
Use as a coordinator before any dispatch - writing a brief, choosing the role and tier, pulling the record, a creative (art, 3D, sound) brief, assigning shared resources or a measurement, requesting a deep-worker session, resuming agents that all died at once, and the daily prune and weekly tool-set upkeep runs. Load skill subagent-dispatch too for the dispatch traps.
shell-and-worktree-traps
by JosiahAllard
Use before running shell commands on the Windows machine (Git Bash or PowerShell), checking line endings, capturing test output, or removing worktrees and branches. Covers silent PowerShell and Git Bash traps that produce wrong or stray results, and the worktree and gate rules workers break by accident.
flaky-test-triage
by JosiahAllard
Use when a load-sensitive or timing-dependent test fails intermittently, or when you must decide whether a change made one better or worse by comparing runs. Covers paired runs and telling a dropped event from a slow one.
subagent-dispatch
by JosiahAllard
Use as a coordinator or any agent about to dispatch a subagent (a kit role or the host's built-in planner or explorer) whose report you will depend on. Covers briefing it, the worktree it writes in, which agents cannot write files, and why a host's output file is not a rescue path for a lost report.
mutation-testing
by JosiahAllard
Use when you write or review a guard/regression test, or build or run a mutation-testing harness (mutate a production line, expect a test to fail). Covers why a guard stays green when it is decorative, stale or never ran, and the harness traps that score a mutant "caught" without testing it.