Handoff
Skill adriannoes/awesome-agentic-ai/cursor-claude-codex/skills/igoruehara-spec-driven/skills/handoff
Use when PAUSING/ending a work session (records the current state in docs/STATE.md to resume later) or when RESUMING (reads docs/STATE.md and the active spec and reassembles the context, proposing the next step). Keeps continuity between sessions of humans and agents. Trigger with /handoff.From its SKILL.md
npx -y skills add adriannoes/awesome-agentic-ai --skill handoffAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
SKILL.md
2.3 KB, 504 tokens by cl100k_base, as published. Nobody here has run it
Translation note: Originally authored in Portuguese (pt-BR) by Igor Uehara (igoruehara/spec-driven, MIT). Translated to English by this hub to keep the repository language consistent. Original content unchanged in meaning; see the upstream repo for the pt-BR source.
Skill: Session handoff (pause / resume)
Keeps project continuity via docs/STATE.md — the volatile working memory
(unlike an ADR, which is a durable decision). Detect the intent from the request or ask.
PAUSE mode (pause / end)
Update docs/STATE.md with the actual state of this session:
- In progress / next step — the active feature/spec and the next concrete action (specific: "implement AC-3 in adapter X", not "continue the feature").
- Recent decisions — what was decided. If it is hard to reverse, create/update the ADR and link it; the STATE only summarizes.
- Blockers — what is blocking and who/how unblocks it.
- Deferred ideas / todos — what was left out on purpose, with the trigger to reconsider.
- Mark the date and author. If there is an open
SPEC_DEVIATION, record it as a blocker. - Offer a commit (
docs: handoff — update STATE) if it is a git repository.
Be concise and actionable. The STATE is for someone (or an agent) to resume cold tomorrow.
RESUME mode (resume)
- Read
docs/STATE.mdand thespec.md/tasks.mdof the active feature mentioned in it. - If a relevant MCP is connected (Jira/Confluence) and the account lock is validated, refresh the context with what changed on the outside.
- Summarize where we left off in a few lines: active feature, last step, open blockers.
- Propose the next step (the "In progress / next step" from the STATE) and confirm with the user before executing.
Rules
- STATE.md is volatile; an ADR is durable. Do not write a structural decision only in the STATE.
- Do not invent progress: faithfully report what was done, what is missing, and what is blocked.
What ships with it
Read from the repository
Just SKILL.md. No reference files, no scripts.