handoff
Standardizes session state transfers through factual, compact artifacts.
Install
mkdir -p .claude/skills/handoff-boshu2 && curl -L -o skill.zip "https://agentskills.codes/api/skills/download/18542" && unzip -o skill.zip -d .claude/skills/handoff-boshu2 && rm skill.zipInstalls to .claude/skills/handoff-boshu2
Activation
This is the description your AI agent reads to decide when to run this skill — the better it matches your request, the more reliably it fires.
Write compact session handoffs. Triggers: "handoff", "write compact session handoffs.", "handoff skill".Key capabilities
- →Write factual session artifacts
- →Include caller-supplied goals and summaries
- →Record completed artifacts and exact evidence paths
- →Document unresolved facts or risks
- →Add optional caller-supplied continuation text
- →Provide best-effort read-only repository identity
How it works
The skill writes a factual session artifact that another context can read, including caller-supplied information and end-state facts.
Inputs & outputs
When to use handoff
- →Transfer work context to new session
- →Document completed coding tasks
- →Summarize findings for team review
About this skill
Handoff
A handoff works because the next context can act on exact paths and facts without trusting the author's memory; any line the reader cannot verify from the artifact itself is decoration, not handoff.
Prompt
Write a handoff for this session in agentops-wt/train2-c: goal was
migrating regen to Go, I finished cli/internal/gates/regen.go and left
scripts/regen-all.sh untouched, tests are green, and the next context
still needs to update docs/CI-CD.md. Write it to .agents/ao/handoff/.
It's working if
Observable in the trace, without reading the prose:
- Every completed item names an exact evidence path like
cli/internal/gates/regen.go, not a chronological narration. - The artifact lands at the caller-named authorized path; new CDLC evidence uses an explicitly selected external non-Git location.
- Readback of that exact artifact path shows the content written.
- Only caller-supplied
continuationtext appears as next action; no owner, tracker state, or verdict is invented.
Contract
Write a factual session artifact that another context can read. Include:
- caller-supplied goal and summary;
- completed artifacts and exact evidence paths;
- unresolved acceptance, findings, causal uncertainties, and risks;
- for a bounded goal, observed native stop/continuation state, measured remaining allowance or explicit measurement gaps, and whether the one helper for the current HOLD incident was already used, with existing evidence references;
- optional caller-supplied continuation text;
- best-effort read-only repository identity when useful;
- the permitted startup association reference, source-store/project/work identity, and observed runtime/session/context IDs or explicit unknowns;
- separately evidenced parent and resume links, observation provenance and cutoff, and any supported work spans with available frozen source bounds and digests, following session associations.
The caller records the work association at dispatch/start and observed native identity at startup through native comments/metadata or runtime facts. Handoff adds end-state evidence; it must not be the first or only work/session link. For an interrupted session, reconstruct only what permitted startup records and native observations support. Missing handoff, missing source, unknown length, unobserved ID or unproved relationship remains explicit; never invent an ID or assign an entire multi-work session to one bead. Preserve earlier unknown/failure observations when later evidence resolves a link.
Check source-owner and recipient/model/destination authorization before reading or copying association metadata. BD/Dolt is versioned and is not a secret store. Use permitted opaque locators where necessary; omit restricted excerpts and sensitive paths from unauthorized destinations. A locator grants no access.
Do not infer a next action, select work, assign ownership, consume the artifact, change tracker or Git state, classify a verdict, govern retries, or restart a runtime. Reading a handoff must not mutate it. A report saying HOLD or NEEDS_OPERATOR is not evidence that a native goal paused; record the actual observed state and operator action still required. Compaction or handoff never resets an allowance, repair bound, or helper incident. Retain informative red and withdrawn claims with their provenance instead of presenting knowledge as monotonically correct.
Named failure mode — optimistic closure: writing "done" for work whose evidence path does not exist, so the next context builds on a phantom.
Anti-pattern: narrating the session chronologically ("first I tried…, then…"). Corrective: record end-state facts — artifacts, paths, unresolved risks — and drop the journey.
Write the artifact to the caller-owned authorized handoff location. For new CDLC memory/episode evidence, require the caller-selected protected external non-Git location; missing routing is a reported gap, with no fallback file in the consumer checkout. Existing requested evidence stays preserved. Standalone non-CDLC handoff retains the explicit requested-proof default below. There is no permanent generic handoff store — an artifact nobody consumes is scratch.
The skill may write Markdown when that better serves a human. The existing
ao session handoff and ao session rehydrate JSON compatibility behavior
below does not itself establish startup associations or an external CDLC route;
use only a command whose actual destination support matches the invocation.
Earlier default compatibility
JSON artifacts already stored under .agents/handoff/ remain read-only
evidence. ao session handoff writes new JSON to .agents/ao/handoff/, while
ao session rehydrate searches both directories and selects the newest
lexical handoff id; if an identical filename exists in both, the canonical
.agents/ao/handoff/ copy wins. No command moves or deletes the legacy files.
Human-authored Markdown consumers receive the exact path, so they do not need
to scan either default. This owning skill contract is the compatibility
authority; no separate migration artifact is required.
Return the artifact path and stop.
When not to use it
- →When a next action needs to be inferred
- →When work needs to be selected or ownership assigned
- →When the artifact needs to be consumed or Git state changed
Limitations
- →Cannot infer a next action
- →Cannot select work or assign ownership
- →Cannot consume the artifact or change tracker/Git state
How it compares
This workflow focuses on recording end-state facts and evidence paths, rather than narrating a chronological session journey.
Compared to similar skills
handoff side by side with the closest alternatives in the catalog.
| Skill | Installs | Updated | Safety | Difficulty |
|---|---|---|---|---|
| handoff (this skill) | 0 | 2mo | Review | Beginner |
| ml-paper-writing | 48 | 7mo | Review | Advanced |
| docs-review | 10 | 8mo | No flags | Beginner |
| claude-md-improver | 21 | 7mo | Review | Beginner |
Try saying
Example prompts that trigger this skill in your AI assistant.
You might also like
ml-paper-writing
davila7
Write publication-ready ML/AI papers for NeurIPS, ICML, ICLR, ACL, AAAI, COLM. Use when drafting papers from research repos, structuring arguments, verifying citations, or preparing camera-ready submissions. Includes LaTeX templates, reviewer guidelines, and citation verification workflows.
docs-review
metabase
Review documentation changes for compliance with the Metabase writing style guide. Use when reviewing pull requests, files, or diffs containing documentation markdown files.
claude-md-improver
anthropics
Audit and improve CLAUDE.md files in repositories. Use when user asks to check, audit, update, improve, or fix CLAUDE.md files. Scans for all CLAUDE.md files, evaluates quality against templates, outputs quality report, then makes targeted updates. Also use when the user mentions "CLAUDE.md maintenance" or "project memory optimization".
markdown-to-html
github
Convert Markdown files to HTML similar to `marked.js`, `pandoc`, `gomarkdown/markdown`, or similar tools; or writing custom script to convert markdown to html and/or working on web template systems like `jekyll/jekyll`, `gohugoio/hugo`, or similar web templating systems that utilize markdown documents, converting them to html. Use when asked to "convert markdown to html", "transform md to html", "render markdown", "generate html from markdown", or when working with .md files and/or web a templating system that converts markdown to HTML output. Supports CLI and Node.js workflows with GFM, CommonMark, and standard Markdown flavors.
write-docs
tldraw
Writing SDK documentation for tldraw. Use when creating new documentation articles, updating existing docs, or when documentation writing guidance is needed. Applies to docs in apps/docs/content/.
marker
benchflow-ai
Convert PDF documents to Markdown using marker_single. Use when Claude needs to extract text content from PDFs while preserving LaTeX formulas, equations, and document structure. Ideal for academic papers and technical documents containing mathematical notation.