CtrlK
BlogDocsLog inGet started
Tessl Logo
SkillPublic

newtask

Prepare an isolated current task checkout using the current host's supported workflow. Check ownership and the base, restore dependencies only when needed, and start the app only for a task that needs it.

Quality

66%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide
SecuritybySnyk

Medium

Suggest reviewing before use

Fix and improve this skill with Tessl

tessl review fix ./.claude/skills/newtask/SKILL.md
SKILL.md
Quality
Evals
Security

newtask — prepare the task checkout

Read root AGENTS.md and the current task checkpoint. Establish repository identity, branch, HEAD, upstream, registered worktrees, existing changes and ownership before writing. Preserve unrelated work. A same-task continuation keeps its established checkout and checkpoint; it does not start over merely because owned changes exist.

Check for overlapping work

Before creating a new branch, check whether the work belongs on an existing owned PR under the repository's bundling policy. When GitHub reads are authorised, inspect open PRs for the actual route, component or surface, rather than only a ledger ID. A missing progress marker does not establish that nobody owns the work. Reconcile confirmed overlap before dependent edits.

The duplicate implementation incident recorded as #292 is historical evidence for this check, not permission to call GitHub. Respect the provider boundary and any armed PR-handoff stop hook. Do not bypass a denied PR read through another transport. git ls-remote and git fetch also contact the remote, require network/authentication and need applicable authority; they are not provider-free fallbacks.

Without an authorised remote read, inspect local branch refs, registered worktrees, current task records and available checkpoints. Record that remote ownership and base freshness remain unverified. Continue independent authorised offline work; do not claim that local branch names prove the absence of another implementation.

Prepare the checkout for the current host

  1. Follow docs/agents/codex-cloud-environment.md for Cloud and docs/agents/codex-desktop-worktree-setup.md for Windows Desktop where applicable. Use the current host's supported task/worktree workflow. A disposable Cloud task checkout does not require a Windows task-start script or another worktree merely because an older local example used one.
  2. For a new local task needing isolation, create a task-specific non-protected branch in an owned worktree. Determine its root from the current checkout and host policy; do not switch or reuse another session's checkout. If an authorised fetch is needed, verify repository identity and fetch origin/main before choosing the base. If working offline, record the actual locally available base SHA and its freshness gap.
  3. Check capacity on the actual destination volume before a dependency install or build. Preserve other sessions' worktrees and services. Low capacity does not grant cleanup authority; use the repository's scoped cleanup workflow only when authorised.
  4. Restore dependencies only when the task needs them and current parity evidence shows they are absent or stale. Inspect the setup helper and coordinator admission first. Where supported, node scripts/setup-codex-worktree.mjs reuses an eligible byte-identical donor before its documented npm fallback. Use the current host's supported restoration route; do not force the legacy Cloud bootstrap or change manifests/lockfiles merely to prepare a checkout.
  5. When relevant, use BASE_FRESHNESS_NO_FETCH=1 node scripts/check-base-freshness.mjs for the default local comparison. The command without that flag fetches origin/main; run node scripts/check-base-freshness.mjs only with already-granted network/provider authority. Distinguish the local tracking ref from a freshly authorised remote observation. If the base moved, assess relevant drift and the task's sync policy; do not discard a continuation or recreate a worktree automatically.
  6. Start or verify the app only for a task that needs it, following AGENTS.md local server safety: run npm run ensure, use its printed URL and verify project identity before browser work. Do not warm unrelated routes or start a watcher during generic onboarding. A documentation-only task needs no app server.

Historical Windows example

On 2026-08-18 the local environment used D:/Repos/PsychSift/.claude/worktrees/ on a ReFS Dev Drive and reported roughly 48–50 worktrees. The then-observed disk capacity, free-space percentage and dependency sizes are historical measurements. They do not identify the current checkout, operating system, capacity or worktree owner. Derive those facts from the current host instead of copying the old path.

Handoff and cleanup

  • Do not stash, reset, clean, relocate or absorb unrelated work.
  • Follow the requested delivery workflow; commit or publish only with applicable authority.
  • Use the handoff skill when available and relevant, without inventing missing tooling.
  • A merged branch does not prove its worktree is disposable. Removal needs verified ownership, current status and the repository's cleanup authority. Do not run periodic cleanup or delete another task's checkout as part of task setup.
Repository

BigSimmo/PsychSift

Path

.claude/skills/newtask/SKILL.md

Last updated
First committed

Is this your skill?

If you maintain this skill, you can claim it as your own. Once claimed, you can manage eval scenarios, bundle related skills, attach documentation or rules, and ensure cross-agent compatibility.