ワンクリックで
continue
Find handoff documents and continue work on a previous task with full context. Companion to the `handoff` skill.
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。
メニュー
Find handoff documents and continue work on a previous task with full context. Companion to the `handoff` skill.
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
This skill should be used when the user asks to "check clean code", "review code cleanliness", "find code smells", "is this function too long", "are these names clear", "review my error handling", "check my tests", or "clean up this code". Also triggers when the user mentions Clean Code, Uncle Bob, Robert Martin, meaningful names, function length, comment rot, F.I.R.S.T. tests, or names a dimension directly ("naming", "boundaries", "concurrency"). Supports auditing all nine dimensions at once or focusing on a single dimension.
Scan Claude Code chat session logs for recurring agent anti-patterns (hedging, stalling, skipping, false completions, preference-asking, dismissing, cosmetic, caveats, flagging) and produce a ranked report with frequency counts, example excerpts, and suggested guardrail rules. Use when user asks to "analyze sessions for anti-patterns", "find bad patterns in logs", "what anti-patterns am I seeing", "agent-guardrails analyze", or wants data-driven guardrail recommendations.
Install agent behavioral guardrail rules into a project's .claude/ directory. Generates a Stop hook bash script and configures settings.local.json. Includes nine battle-tested rules (no-guessing, no-stalling, no-preference-asking, no-false-completion, no-skipping, no-dismissing, no-cosmetic, no-caveats, no-flagging). Use when user asks to "install guardrails", "set up agent guardrails", "add behavioral hooks", "agent-guardrails install", or wants to enforce assistant discipline.
Test installed agent guardrail hooks by sending trigger phrases through the stop hook script. Checks hook installation first, then runs two test phrases per rule. Use when user asks to "test guardrails", "verify hooks", "check guardrail patterns", "agent-guardrails test", or wants to confirm hooks are working.
Re-analyze Claude Code session logs against existing agent guardrail rules to measure effectiveness, catch false positives, identify missed anti-patterns, and refine regex patterns. Use when user asks to "update guardrails", "refine guardrail rules", "check guardrail effectiveness", "agent-guardrails update", "tune guardrails", or wants to improve existing behavioral rules based on real usage data.
Use when the user wants to write a "product spec" or "product specification", frame a "product bet", or define a falsifiable hypothesis before building. Also triggers on "replace the PRD", "retire the PRD", "the PRD is too long / too vague", "tighten this spec for an engineer or agent", "what's the bet", "what problem are we solving", "acceptance criteria", "success criteria", "kill / scale / graduate thresholds", "measurement plan", or when scoping a feature's problem, bet, success behaviors, and evaluation before any behavioral spec. Part of the spec-dd workflow; precedes /spec-dd:spec.
| name | continue |
| description | Find handoff documents and continue work on a previous task with full context. Companion to the `handoff` skill. |
| argument-hint | Optional filter — topic keyword to narrow candidates. |
| disable-model-invocation | true |
Find handoff documents written by the handoff skill, present them to
the user, then execute the continuation protocol specified in the chosen
handoff.
Look in docs/handoffs/*.md, relative to the current working directory.
If the directory does not exist, tell the user:
"No
docs/handoffs/directory found here. If you're in the wrong working directory, change to the project root and run continue again. If you have a handoff elsewhere, share its path and I'll read it directly."
If a filter argument is provided, match it case-insensitively against the H1 topic and the Goal section of each handoff.
For each handoff file, extract:
# Handoff: <topic>).## Goal.## Next agent: start here.## Metadata > Timestamp:.## Metadata > Mode:.## Metadata > Verification:.## Metadata > Prior handoff:.If a file is malformed, skip it but report "Skipped malformed:
()" so the user knows it exists. Treat as malformed if any of
these are missing: H1 starting with # Handoff:, ## Goal section,
## Next agent: start here section, ## Metadata section.
Zero candidates. Tell the user honestly:
"No handoff files found in
docs/handoffs/. If you have a handoff elsewhere, share its path and I'll read it directly."
Do not invent options.
One candidate. Don't make the user pick from a list of one. Show the parsed preview and confirm:
"Found one handoff:
(, ) Goal: First action: Path:
Continue this? (yes / no)"
Multiple candidates. Sort by timestamp, newest first. Show as a numbered list with parsed preview for each:
"Found N handoffs. Newest first:
(, ) Goal: First action:
(, ) ...
Which would you like to continue? (number, or 'none')"
Flag any handoff with Verification: non-interactive using a
[unverified] marker — its Intent and Stance were inferred without
user confirmation.
Flag any handoff older than 30 days with a [stale] marker. Do not
hide it; the marker invites scrutiny.
Once the user has chosen a handoff:
"This handoff's Intent and Stance were inferred without verification. Before I act on them, can you confirm or correct them? Here's what was inferred:
"
Leave the handoff file alone. Do not delete, move, rename, or modify it. The user manages handoff files; this skill only reads them.
If the user passes an argument (e.g., continue auth refactor), treat
it as a filter against topic and goal. If exactly one handoff matches,
go directly to the single-candidate confirmation. If multiple match,
show the filtered list. If none match, fall back to showing all
handoffs and tell the user the filter matched nothing.