Continues interrupted Claude Code work after local history is read and reconciled — reconstructs the original outcome, unfulfilled requests, and the next action. Use when given a Claude Code Session ID, asked to pick up prior Claude work, told a run was interrupted or compacted, or asked to take over without `claude --resume`. Not for Codex sessions (use continue-codex-work).
74
93%
Does it follow best practices?
Run evals on this skill
Adds up to 20 points to the overall score
Passed
No findings from the security scan
Continue means execute the still-unfinished business task. It does not mean summarize the transcript, perfect the history tool, or start a new experiment.
Invoke daymade-claude-code:read-claude-code-history for the exact Session. Ask
that Skill for a chronological evidence briefing, not separate “last user” and
“last assistant” buckets. If the Session ID is unknown, use its search/inventory
route first.
Do not act until the receipt states:
If a clipped section could contain the objective or a correction, re-read it with the reader's full mode. If identity, ancestry, or authorship is unresolved, stop; continuing from a guessed Session is worse than asking for the missing evidence.
Before the first project-changing tool call, write this compact internal contract from evidence. Do not invent a value to fill a blank.
| Field | Evidence required |
|---|---|
| Original business outcome | Earliest still-governing human request, not a later implementation detail |
| Current explicit request | Latest human request that is not merely “继续” |
| Already completed | Independent current-state verification, not the old Agent's claim |
| Still unfulfilled | Requested result with no completion evidence |
| User corrections / do-not-repeat | Human messages that rejected a route, output, or assumption |
| Proven assets and successful routes | Existing code, docs, Skills, outputs, or commands that previously worked |
| Next direct action | One action whose success visibly reduces the unfulfilled business result |
When the latest message is only “继续,” inherit the objective from the verified chronology. Never treat the cue itself as the specification.
AGENTS.md/CLAUDE.md and authoritative
project state; historical instructions may be stale.Take the next direct action from the contract and verify its result. Skill edits, review expansion, documentation systems, and infrastructure cleanup remain subordinate unless the user explicitly made one of them the business outcome or they are strictly required to unblock it.
Before changing direction, ask mechanically: “If this succeeds, which original unfulfilled result becomes smaller?” If the answer is none, record it as a later proposal and return to the task.
Do not run claude --resume or claude --continue. Do not overwrite unrelated
working-tree changes.
Do not say “continued successfully” when only context extraction or a process step completed; the business result is the completion unit.
d8d8528
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.