- Home
- Skills
- Technical Documentation
- Project Handover: a dossier your successor can trust
Works with the AI Agents you already use
Project Handover: a dossier your successor can trust
They fail on the confident, wrong page: an architecture that no longer matches the code, a deploy step that changed, a dependency removed months ago.
$12
Project Handover: a dossier your successor can trust
Example session with this skill installed
I'm handing this project over to a new maintainer next month. Write the full handover dossier.
- Read your context and instructions
- Compiled the project handover
- Generated the document
Summary
I wrote the full dossier in docs/handover/ (7 documents, about 900 lines) from the repository at commit e3d2b0d, and exported it to Word, with the sign-off also as PDF. 46 points could not be verified from the code; they are marked TODO (human) and gathered into 25 questions for you in the README.
Documents
| # | Document | What it answers | Status |
|---|---|---|---|
| 01 | Functional | What the plugin does, for whom, and what is live | draft, 3 open points |
| 02 | Technical | Architecture, stack, integrations, state, and seven informal ADRs | draft, 2 open points |
| 03 | Inventory | Repos, distribution channel, accounts and secrets by reference | draft, most open points are ownership |
| 04 | Runbook | How to release, set up a machine, configure, and what not to touch | draft, tests not run for this draft |
| 05 | Pending work and risks | Backlog, known issues found while reading, dependencies, dates | draft |
| 06 | Handover sign-off | Signable checklist, delivery channel, rotation, ownership | draft, to fill on the day |
What the reading turned up
Three things are worth knowing before anything else, because they are wrong or misleading today. The /jev-doctor command most likely runs a path outside the plugin since the plugin root moved (commit f4656fd). The README says every done is re-checked, but in local mode it is not (README.md:120 vs server.py:267). And two in-repo sources give different step times for local mode (server.py:220 says 2 to 4 s, SKILL.md:45 says 5 to 13 s). Details and fixes are in 05-pending-risks.md.
Open points for the outgoing owner (excerpt)
- 02: Why monkeypatch the upstream library instead of forking or contributing upstream
- 03: GitHub settings not visible from a clone (branch protection, collaborators, secrets)
- 03: Confirm the
TEXT_MODEL_API_KEYdefault inrun.sh:19is a harmless placeholder - 04: Confirm tests and lint pass on the handover date
- 04: Mirror or backup of the GitHub repository
- 05: Which local-mode step time is right (2 to 4 s or 5 to 13 s)
- 06: Names, organisations and contacts of both parties, and the handover date
Exported
docs/handover/export/: README, 01 to 06 as .docx (3 Mermaid diagrams embedded as images) and 06-handover-signoff.pdf (A4).
Next steps
- Answer the 25 open questions in the README; each answer goes into the document it came from.
- Add docs/handover/export/ to .gitignore unless you want the Word and PDF files versioned.
- On the handover day, fill in 06, deliver secrets through the agreed channel and work through the rotation checklist.
06-handover-signoff.pdf
PDF · document
Example file from a real run - the skill writes it into your workspace.
Connects securely to your tools. The creator never sees your data.
About this skill
Handovers rarely fail for lack of pages. They fail on the confident, wrong page: an architecture that no longer matches the code, a deploy step that changed, a dependency removed months ago. The person taking over trusts it, acts on it, and breaks something.
This skill has your agent write the handover dossier from facts it checks in the repository and whatever infrastructure you give it, and cite each one (path:line, commit, command). Anything it cannot verify, such as who has access, production hosts or the reasons behind old decisions, becomes a TODO (human) for you to answer. You get an honest scaffold with every gap in plain sight, not a polished document nobody can trust.
What it writes (in docs/handover/, Markdown as the source of truth):
- README: index, state of each document, and the list of open questions for the departing owner.
- 01 Functional: what the product does, for whom, and what is in production versus in development.
- 02 Technical: C4 context and container diagrams in Mermaid, stack with real versions, APIs and integrations, data model, and informal ADRs (context, decision, why, rejected alternatives).
- 03 Inventory: repos, environments, servers, domains and certificates, SaaS accounts, and secrets by reference only.
- 04 Runbook: deploy, per-environment variables (names, never values), dependencies, backup and restore, monitoring, recurring tasks, and the tribal knowledge.
- 05 Pending work and risks: backlog with enough context to pick it up, known issues, critical third parties, and every date that must not slip.
- 06 Sign-off: a signable record with the secret delivery channel, a credential rotation checklist and ownership transfers.
Export: one command turns the dossier into Word (.docx) and the sign-off into PDF (A4 or Letter), with the Mermaid diagrams rendered as images. Bring your own Word reference document for your brand's styles.
Secrets stay safe: it never writes a secret value. It lists names and paths in your secret manager (Vault, 1Password, AWS Secrets Manager and others), treats look-alike defaults as secrets, and puts every shared credential on a rotation checklist for after the handover.
Writes like a person: prose that explains how things connect and why, with lists and tables only for what is genuinely enumerable. Dates are absolute, status markers are consistent, and every document opens with a metadata block.
You can ask for the whole dossier or one block: just the runbook, just the inventory before rotating credentials, or just the technical document.
Tested on a real Python project: the agent produced a 7-document dossier (about 900 lines) with 46 open questions for the owner, and along the way caught a broken command path, a README claim the code doesn't keep, and two docs that disagreed with each other.
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Technical Documentation.
It can only state what it can check in the repository and in the infrastructure you give it access to. Production hosts, who has access, contracts and deadlines that live in someone's head come out as TODO (human) items for you to answer. It doesn't run deploys or rotate credentials; it lists what to rotate. Word/PDF export needs pandoc.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before buying a skill.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 1 day ago
- Software
Gitpandoc (for Word/PDF export)mermaid-cli (mmdc), if the dossier has diagrams
- Optional
- Google ChromeorChromium (PDF and diagram images)
The dossier itself is plain Markdown and needs nothing. Export needs pandoc, plus mermaid-cli when there are diagrams; PDF and diagram images also use Chrome. The export script never downloads or installs anything: if a tool is missing it stops and says how to install it.
Permissions
Discovery is read-only. It writes only docs/handover/ in the project you point it at. It never writes secret values; it lists secrets by name or path in your secret manager. Nothing is downloaded or sent anywhere.