handover
Skill27 starsChanged 18 days agoApache-2.0View on GitHub
Produce a structured handover note when the context window fills or the user asks to wrap up. Take the real percentage from /context or, in the Claude desktop app, the get_usage tool; never estimate, and where neither exists record the gap verbatim rather than inventing a figure. Triggers when the context band reaches amber or red, when the user says "wrap up", "handover", or "hand off", or before deliberately ending a long working session.
51 other projects use a version of this.
Adopted by
Most recently taken up by
- thiagobluhm/o1memtheir own version, first seen Nov 8, 2026
- hmislk/hmistheir own version, first seen Nov 2, 2026
- loonylabs-dev/claude-setuptheir own version, first seen Nov 1, 2026
- oubakiou/skillstheir own version, first seen Nov 1, 2026
- vrppaul/claude-reviewtheir own version, first seen Nov 1, 2026
- konlue/skillstheir own version, first seen Nov 1, 2026
Used with
Projects with it also use harness-init, incident and 8 more.
Other files people use alongside this one.
Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
steptianSkillharness-init
初始化 .harness 开发框架,集成代码修改影响范围评估和测试覆盖建议功能。触发词:'初始化新项目'、'harness init'、'更新 harness'、'分析影响范围'、'评估修改影响'
Also in 3 projects that use this
JkudjoSkillincident
Production incident response — structured triage, hypothesis tree, safe remediation, incident record
Also in 3 projects that use this
nicolasestremSkillbrainstorm
Structured idea generation and solution exploration with systematic analysis.
Also in 3 projects that use this
GoogleSkillto-tickets
Break a plan, spec, or the current conversation into a set of tracer-bullet tickets, each declaring its blocking edges, published to the configured tracker — edges as text in one file per ticket locally, or native blocking links on a real tracker.
Also in 3 projects that use this
BlockSkillplan
Draft a TDD, PRD, or lightweight plan in the conversation without committing it.
Also in 3 projects that use this
MicrosoftSkillcaveman
Ultra-compressed response style that reduces output token count while preserving technical accuracy, with intensity levels and auto-clarity safety rules
Also in 3 projects that use this
Other versions51 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 51
- me2resh/apexyard★ 542same name
- hmislk/hmis★ 244same name
- chenklein26-maker/Harness-Starter★ 119same name
- k1ein-chen/Harness-Starter★ 119same name
- eai-org/agent-toolkit★ 85same name
- manusco/resonance★ 38same name
- chacosoldier/compabob★ 34same name
- L-X-T/ng-agentic-skills★ 13same name
- marcogalluccio/company-brain-os★ 7same name
- Cor-Incorporated/claude-code-skills★ 5same name
41 more not listed
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository56 other files it gives its agents.
- ai-surface-disciplineSkill
- blast-radius-grepSkill
- checkpoint-logSkill
- commission-the-rosterSkill
- committee-reviewSkill
- confirm-before-pushSkill
- constant-time-secret-compareSkill
- context-economySkill
- contract-reviewSkill
- cross-agent-reviewSkill
- date-parse-utc-safeSkill
- db-migration-verificationSkill
- deliverable-integritySkill
- dependency-upgrade-verificationSkill
- earn-every-lineSkill
- email-delivery-verificationSkill
- enforce-invariants-in-buildSkill
- env-change-verificationSkill
- findings-are-evidenceSkill
- fixtures-from-realitySkill
- flag-deferred-itemsSkill
- guard-the-spend-pathsSkill
- honest-failure-surfacingSkill
- legal-citation-integritySkill
- legal-notice-handlingSkill
- live-data-surgerySkill
- live-state-firstSkill
- lock-at-the-chokepointSkill
- mass-red-triageSkill
- measure-per-routeSkill
- no-silent-data-dropSkill
- one-real-rideSkill
- outbound-side-effect-idempotencySkill
- parallel-work-reconSkill
- pin-to-what-was-seenSkill
- plan-firstSkill
- price-the-spendSkill
- prove-it-can-failSkill
- reproduce-the-real-buildSkill
- rerun-before-verdictSkill
- safe-smokesSkill
- secrets-in-outputSkill
- server-side-authoritySkill
- skill-library-builderSkill
- spawn-from-declared-stateSkill
- statute-to-obligations-registerSkill
- substantiate-outward-claimsSkill
- team-loopSkill
- tool-layer-bytesSkill
- trace-one-recordSkill
- unordered-eventsSkill
- unslop-codeSkill
- unslop-textSkill
- unslop-uiSkill
- verified-citationsSkill
- verify-the-effectSkill
The file
We split this skill into the rules it gives an agent
3 rules. 0 are common, 2 are distinctive, and 1 is very specific to this project.
0
Common
Most of the 38 edited versions we read keep these.
2
Distinctive
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 18
Run /context first and use the real figure.
- Distinctive: line 44
Deferred items: Every item deferred this session, each with its grep anchor, per the flag-deferred-items skill.
- Distinctive: line 56
Every diagnosis and causal claim in the note carries a confidence marker: verified, with the command or evidence that proved it, or …
“Common” is measured: kept by at least 6 in 10 of the 38 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Handover note
The global operating rules define context bands: green below 50%, yellow 50 to 65%, amber 65 to 75%, red above 75%. At yellow, mention the reading and keep commits small with explicit scope cuts. At amber, take no new work units. At red, stop and hand over. This skill defines what the handover contains and how to produce it.
When this applies
- The context band reaches amber or red.
- The user says "wrap up", "handover", "hand off", or similar.
- A long session is ending deliberately and the next session needs to resume cleanly.
Step 1: measure, never estimate
Run /context first and use the real figure. Do not guess or estimate the percentage. Run /context and use the figure it returns. The whole point of the bands is that the decision to wrap is made on a real reading, not a feeling.
In the Claude desktop app the agent cannot type /context, but the reading is one tool call
away: mcp__ccd_session_mgmt__get_usage with session_id: "self". It is usually deferred, so
load it through ToolSearch first. Its context.percentUsed is the reading; the same call reports
the plan's 5-hour and weekly limits, which matter when several sessions run at once. On
2026-09-24 a desktop session wrote the gap line below into a handover while this tool sat in its
deferred list, and the user had to supply the 38 % reading. Search the deferred tools for
"context usage" before declaring a gap.
Where neither /context nor a usage tool is available (a non-interactive orchestrator or
headless session), an honest gap is the reading. Write "no reading: /context unavailable in this
harness" verbatim into the note's wrap-up context, never an invented or estimated
percentage, and make the band decision from the harness's own signals (context-compaction
warnings, summarisation system reminders). Five consecutive sessions on one project
(2026-07-31 to 2026-08-12) wrote that gap into prose and re-declared this rule change due
before the rule caught up with the harness it actually runs in.
Step 2: the handover note
A fixed shape, so the next session can read it once and resume:
- Session goal. One line: what this session set out to do.
- Branch and worktree. The working branch, and the worktree path if non-standard. The next session starts on
mainby default and otherwise has to rediscover this fromgit log. - What landed. Commits and PRs with sha and a one-line description each. Mark each as verified or unverified.
- In flight. What is half-done, with exact
file:lineranges, not bare paths. Enough that the next session can open the file at the right place. - Deferred items. Every item deferred this session, each with its grep anchor, per the flag-deferred-items skill. The note records where the anchor was planted.
- Verification still outstanding. Migration catalog queries to run, smokes to run, post-rollout checks not yet done.
- Blockers and open questions. Anything waiting on a decision or an external answer.
- Next actions. Ordered and concrete. Each is a single first step, not a theme.
- Traps and working agreements. Project-specific gotchas surfaced this session, plus standing rules that bit or nearly bit (confirm before push, money is pounds not pence, grep the live schema rather than trusting handover docs).
Dates are absolute, not relative. Money is stated in pounds.
Where the project keeps a checkpoint log (per the checkpoint-log skill), synthesise items 3, 4, and 5 from the log rather than from memory; the log is the record, the handover is the digest.
Every diagnosis and causal claim in the note carries a confidence marker: verified, with the command or evidence that proved it, or unverified. The note opens with one line: "diagnoses in this note are unverified unless marked". A recommendation built on an unverified diagnosis inherits the marker. The incident behind this (2026-07-20): a second-order claim about a fix, reasoned from structure rather than probed, was written into a migration header and DECISIONS as fact, and a reader had no way to tell asserted from proven until a smoke forced the correction.
Step 3: persist it
Write the note to the project's handover location, or to DECISIONS.md / LESSONS_LEARNED.md where an item qualifies for those. If no handover location is established, write to HANDOVER.md in the project root. Append; do not overwrite. The history of handovers is itself useful, and an overwrite loses the trail. If something this session qualifies as a documented decision or a lesson learned, prompt to record it.
In a project that keeps team/NOW.md (the team-loop skill), every session end overwrites
NOW.md from team-loop/templates/NOW.md, at most 40 lines: item 1 as its topic:, item 8's
first step as next:, and items 2 to 6 and 9 under its headings, with landed commits still
marked verified or unverified and its branches: and ask: lines current. Git keeps the
trail, so the overwrite loses nothing. Item 7's open questions go to the project's Ask queue,
not into NOW.md. That replaces the appended note for a routine end, amber-band wrap-ups
included. An unusual end also gets a dated docs/HANDOVER_<topic>_<date>.md in the full shape
above: the red band, a crash or forced stop, a move to another machine, or a scheduled run
stopped by a usage limit. Steps 1, 2 and 4 apply either way.
Then get the note committed before the session ends, with the user's per-action yes. A
note left untracked reaches neither the other machine nor the next reader's git log,
and it is invisible to the session that resumes from a different checkout. On 2026-09-14
four handover and review notes from three sessions were found untracked in this
library's docs/, one of them already superseded.
Step 4: promote before closing
Sweep the note's traps and working agreements before finishing. Anything that would change how a future session behaves is a lesson, not a handover line: promote it to the project's LESSONS_LEARNED.md now, with its misses-log line, or plant a grep-able FORWARD anchor at the landing point (flag-deferred-items). Knowledge that stops in a handover never reaches the library's admission pipeline. The incident behind this (2026-07-27): five coordination rules learned the hard way in one session sat in a handover section behind "consider folding these in later", and were never folded in.
What this skill does not do
It does not decide to wrap up; the context bands do that. It does not push or commit on its own. It produces the note and persists it, leaving the push gate intact.
Why
Long sessions fail at the seams: the next session re-discovers what the last one already knew, or misses a deferred item or an unrun verification. A fixed handover shape, produced from a real context reading, makes the seam cheap to cross and keeps deferred and unverified work from falling through it.
Additions from cross-repo lessons (ratified 2026-07-23)
- Stamp superseded sections. A session that makes a prior handover's "next build" section stale stamps that section superseded in the same PR — a one-session-stale handover cost real rework.
- Carry-forward items cite a live-state verification or are marked unverified. A "still open" item once rode four consecutive handovers a month after the work was done. Re-verifying at write time is cheaper than the duplicate build it prevents (see live-state-first).
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

