メニュー
Use when work was interrupted, state looks stale, or user says continue/where were we. Not for brand-new work.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
Use when a plan or concrete tasks exist and user said build/go/implement. Not to invent scope or claim done without verify.
Use when goal/spec is clear and you need ordered tasks before writing code. Not while still clarifying.
Use when about to claim done/works/fixed, or user asks test/check/prove. Run evidence; not for pure planning chat.
Use when another installed plugin/skill may do the job better, or to route specialists. ACC stays workflow owner.
Use when user wants an explicit recorded decision or blocker saved to artifacts. Not silent monitoring.
Use when request is too vague to plan or build. Not when goal and constraints are already clear.
| name | resume |
| description | Use when work was interrupted, state looks stale, or user says continue/where were we. Not for brand-new work. |
Script root: ACC_PLUGIN_ROOT from SessionStart (hooks inject). PLUGIN_ROOT is hooks-only.
Use $resume when a session was interrupted or the state looks stale.
Reply rule:
If still on Codex: try native first — codex resume / --last, or codex exec resume,
or app-server thread/resume. ACC portable file is for work truth across tools/models.
resume: continue from intact statereconstruct: rebuild state from artifactsrepair: fix stale or corrupt workflow stateabandon: start a fresh run while preserving artifactsPortable handoff first (any tool):
.codex/anyone-can-code/artifacts/PORTABLE_HANDOFF.md
Goal, plan, next step, memory pointers, optional session_id.
Run python3 "<ACC_PLUGIN_ROOT>/scripts/runtime_info.py" --resolve-project "."
Use returned project_root. If result is ambiguous, block recovery
and show candidates; never merge or replace competing state automatically.
Canonical
.codex/anyone-can-code/state/workflow.json. Confirm derived files carry same
transaction ID. Use history, ledgers, artifacts, and AGENTS.md only as
supporting evidence.
Use scripts/canonical_state.py recovery behavior to re-anchor from canonical
state. Read active_task_capsule for goal, task, decisions, boundaries,
evidence, and next action. Before known restart, handoff, or compaction risk,
save a context-transition capsule when possible.
If usage_checkpoint or session evidence says primary usage is at least 85%,
checkpoint before continuing. At 90%, split before more work. At 94%, stop now
unless the user explicitly chooses to continue.
Ignore legacy workflow truth fields if present. Resume from canonical
verification, transaction, capsule, recovery, and next-action fields only.
Choose the lightest recovery mode that is still safe. If active truths conflict, block work and repair state before resuming. Report every missing or uncertain field. Never invent recovery context. Hooks may help with session context, but resume must work when hooks are absent, failed, or circuit-broken. Do not treat hook output as canonical truth.
"What did we decide about X?" — do not shrug. Run
python3 "<ACC_PLUGIN_ROOT>/scripts/past_answer.py" "<keywords>" and answer with
the date of each record. Nothing found: say "no record" honestly.
Next: continue with $plan / $execute / $verify / $fix from recovered state.
When this skill's job is finished: