codewhale-hq / cw-handoff

cw-handoff

A handoff is read by an agent with none of your context and every incentive to
believe you. That makes an optimistic handoff worse than no handoff: it converts
your guesses into the next session's premises. Write it so the reader can
re-derive the state instead of trusting the prose.

Stage 6 of the loop, and the one that makes the loop a loop: the next session
starts at [cw-orient](../cw-orient/SKILL.md) with what you leave here.

When to use

• Ending a session with work in flight.

• Asked for "a prompt for another agent", branch takeover instructions, a
  continuation note, or a summary of current state for async work.

• Handing a lane to a different model, a fleet worker, or a remote session.

Workflow

1. Write it as a prompt the next agent can paste directly. Not a report
   about the work — instructions for continuing it.

2. Open with the refresh block, not with your summary. The first thing the
   reader should do is verify you:
   `bash
   cd <repo path>
   git status --short --branch && git branch --show-current
   git log --oneline --decorate -20
   git worktree list
   ./scripts/release/check-versions.sh
   `
   Tell them to trust that output over anything below it.

3. Cover these, in whatever order fits; drop sections that are empty:

   • Repository path and expected branch or worktree.

   • The authority line: what they may and may not do without asking. Default to
     local-only — no push, merge, tag, publish, GitHub Release, or destructive
     cleanup without explicit approval.

   • Durable files to read, ordered by importance: the scoped AGENTS.md for
     the area, then the specific docs (docs/CACHE.md,
     docs/MOTION_CONTRACT.md, docs/ARCHITECTURE.md, crates/tui/AGENTS.md)
     the task actually touches.

   • Commits already landed, with SHAs.

   • Dirty worktree caveats, naming uncommitted files explicitly, and whose
     they are. This is the single most useful line in most handoffs.

   • The next slices, in priority order, each bounded the way
     [cw-slice](../cw-slice/SKILL.md) bounds one.

   • The verification gate for those slices — the smallest correct one from
     [cw-gates](../cw-gates/SKILL.md), plus any known-flaky names.

   • Open decisions that genuinely belong to Hunter.

4. Separate three states, and never blur them: done and verified (with the
   command output that proves it), suspected (a hypothesis, labeled), and
   blocked (with what unblocks it). If you did not run it, it is not done.

5. Say what the branch is for. Local-only, pushed for backup, or intended to
   stay unpushed — the next agent cannot tell from git alone, and guessing
   wrong is how work gets force-pushed away.

6. Record missing external receipts. If the task was local-only, say which
   evidence you could not gather (CI state, registry state, review threads)
   rather than inferring it. A named gap is useful; a confident guess is not.

Red flags / don't

• Don't imply publication happened unless you verified registry, tag, or release
  state live.

• Don't hand off a narrative. Prefer concrete paths, commands, and SHAs.
• Don't summarize away the dirt. Unnamed uncommitted files get destroyed.

• Don't hand the next agent decisions you could have made. Reserve Hunter-facing
  choices for product direction, irreversible actions, and visual judgments that
  need eyes.

• Don't copy a previous handoff's state forward. Re-derive it; that is what
  step 2 is for.

• Don't include secrets, tokens, or provider credentials in a handoff file.

Output

A single paste-ready block with the default shape — refresh commands,
authority line, files to read, landed SHAs, named dirty files, prioritized
next slices, the verification gate, and the open decisions — with done /
suspected / blocked visibly separated. Compress when trivial, but never
compress away the state separation: that is what keeps guesses from
becoming the next session's premises.