菜单
Append a Four-Noble-Truths-shaped task entry to task-log.jsonl (Dukkha → Samudaya → Nirodha → Magga). Use when the user says "log task", "task-log", "record this task", "append to task log", or at the start of any non-trivial task per the AGENTS.md protocol.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
Verify translation parity for BWOC docs (English is canonical). Reports any docs/en/*.en.md without a matching translation in each present docs/<lang>/ (th today, any BCP 47 / ISO 639-1 language in future), and — inside a git repo — any EN file changed in the current diff whose translations were not also changed. Use when the user says "check bilingual", "parity check", "EN TH", "check translations", or before a PR touching docs.
Audit every *.md file in the repo against the unified naming standard in docs/en/NAMING.en.md. Reports deviations by category (root-level OSS standard, spec docs, notes, etc.). Use when the user says "check naming", "audit md naming", "find naming violations", or before a release. Read-only.
Run the backend-neutrality audit on the agent template or a specific incarnated agent. Reports hardcoded model IDs, vendor-specific phrasing, missing symlinks, and YAML/wikilink violations in AGENTS.md. Use when the user says "check neutrality", "neutrality audit", "verify backends", or before a PR touching AGENTS.md.
Scaffold a new BWOC agent by running the template's incarnate.sh, then guide the user through filling in the {{camelCase}} placeholders in AGENTS.md and config.manifest.json. Use when the user says "incarnate", "new agent", "clone the template", or names a new agent to create.
基于 SOC 职业分类
| name | task-log |
| description | Append a Four-Noble-Truths-shaped task entry to task-log.jsonl (Dukkha → Samudaya → Nirodha → Magga). Use when the user says "log task", "task-log", "record this task", "append to task log", or at the start of any non-trivial task per the AGENTS.md protocol. |
| disable-model-invocation | true |
User-triggered. Side effect: appends one line to task-log.jsonl (append-only — never rewrite the file).
$ARGUMENTS — free text describing the task. The skill extracts or asks for the four fields below.
task-log.jsonl. Default to the current working directory's root (a real incarnated agent). If absent and we're inside the framework root, ask the user whether to create one here (the framework itself does not normally log tasks) or to point at a specific agent path.modules/agent-template/AGENTS.md §2.2:
{
"taskId": "TASK-NNN",
"moduleName": "...",
"branchName": "feature/TASK-NNN",
"worktreePath": "/tmp/<agentId>/TASK-NNN",
"status": "in_progress",
"startedAt": "<ISO-8601 UTC>",
"lastAction": "logged via /task-log",
"completedAt": null,
"blockedReason": null,
"dukkha": "...",
"samudaya": "...",
"nirodha": "...",
"magga": "..."
}
taskId: increment from the last line of the file (tail -n1 task-log.jsonl | jq -r .taskId). If the file is empty, start at TASK-001.startedAt: date -u +%Y-%m-%dT%H:%M:%SZ.moduleName, worktreePath, agentId: read from config.manifest.json if present; otherwise ask.jq -c . before writing to ensure no malformed entry corrupts the file.
echo "$ENTRY" | jq -c . >> task-log.jsonl
taskId. Do not echo the whole file.sed -i, >, or otherwise rewrite existing lines.task-log.jsonl by design; this skill is primarily for incarnated agents.Name Sammā-sankappa (Right Intention) — the four-truth structure forces intent to be made explicit before action.