status

name	status
description	Use when user wants a checkpoint: where we are, next step, blockers. Not for deep rebuilds.

Status

Use $status for a compact checkpoint.

Reply rule:

• talk strict caveman only
• keep answer short

Script root: use ACC_PLUGIN_ROOT from SessionStart context (hooks inject it). PLUGIN_ROOT is hooks-only — do not assume it in the agent shell. Run python3 "<ACC_PLUGIN_ROOT>/scripts/runtime_info.py" --resolve-project "." before reading state. Use returned project_root. If result is ambiguous, stop and show candidates instead of choosing or reporting root state.

Read canonical .codex/anyone-can-code/state/workflow.json first. state-current.md is derived and must carry same transaction ID. Treat canonical verification, active_task, next_action, recovery, and transaction_id as truth. Ignore legacy truth fields such as status_line, work_state, verification_state, states, top-level evidence, and top-level unverified if they appear.

Also read:

• .codex/anyone-can-code/artifacts/VERIFICATION.md if present
• .codex/anyone-can-code/state/install.json
• source manifest if repo has .agents/plugins/marketplace.json
• installed cache manifest if found under ~/.codex/plugins/cache/...

Return:

• status from canonical verification and tasks, using only closed words: in scope, designed, approved, implemented, verified, blocked, deferred
• active task
• next action
• project root
• source version
• runtime version
• memory mode: portable-markdown or fallback
• memory storage health and latest migration/import receipt status
• viewer mode: none, obsidian, or future acc-viewer unavailable
• failures
• silent failures
• unverified work
• remaining uncertainty
• blockers or missing evidence
• evidence levels checked, especially interaction test, visual QA, and user acceptance when user-facing quality is being discussed
• active-task capsule saved time and next action
• canonical transaction ID and any derived-view disagreement
• legacy truth fields found, if any, as a repair warning
• command_guard state when command, package-manager, browser, server, Git, or tool work is active
• Windows command safety: npm.cmd required in PowerShell, Bash-only || avoided, and repo root resolved before Git commands
• failed command attempts that still need visible recovery
• usage_checkpoint state: below 85% can continue, 85% needs checkpoint, 90% needs split, and 94% needs stop-now unless the user explicitly chooses to continue
• patch_retry state: failed patch count, whether exact target reread happened, whether retry is allowed, and whether stop/replan is required after repeated failed patch attempts
• mechanics_docs_gate state: whether work touches platform mechanics, whether a docs brief exists, whether controlled proof has recorded uncertainty, and whether code changes are blocked

Never use done, complete, finished, or similar words as workflow states. Example: Status: build implemented, tests verified, deploy blocked. Never report works, proper, perfect, or user accepted unless matching interaction, visual QA, or user-acceptance evidence is recorded.

Live helpers

Desktop package (CLI = separate plugin). Cheap truth scripts when useful: host_detect.py --guidance, loop_registry.py, ai_observability.py, past_answer.py.

Done — back to normal

When this skill's job is finished:

1. Stop following this skill.
2. Reply short and normal (caveman).
3. Do not keep this workflow for the whole session unless the user asks again or a new skill matches.