- Home
- Skills
- Agents & Orchestration
- Session Handoff Brief
Works with the AI Agents you already use
Session Handoff Brief
Turns the agent session you are in into a handoff document a different session can resume from without having seen any of it
$5
/moSubscription · cancel anytime
Secure checkout via Stripe
- Always the latest version
- Delivered live to your agent via MCP
- Cancel anytime, access ends at period end
Session Handoff Brief
Example session with this skill installed
I'm stopping here. Write HANDOFF.md so whoever picks up INFRA-88 next isn't starting from zero.
- Read your context and instructions
- Compiled the session handoff brief
- Generated the document
Written to HANDOFF.md at the repo root, untracked and uncommitted.
Handoff - INFRA-88 staging state to S3 - 2026-09-03
Covers: this session's work on INFRA-88, staging only | Not covered: prod migration, the modules/backend extraction, any change to the resources declared in envs/staging/main.tf
Resume here
Make a copy of envs/staging/terraform.tfstate somewhere outside the repo and note the path in this file - the copy this session said it made does not exist on disk. Do that before running terraform init -migrate-state, which is the step that rewrites the local state. File to open: envs/staging/terraform.tfstate.
State on disk
Branch: infra-88/remote-backend, 1 commit ahead of main (ebabf68 "Placeholder for a shared backend module"; the only diff against main is modules/backend/README.md).
Uncommitted: envs/staging/backend.tf - untracked, never committed, never used by a terraform run. Contains the full S3 backend block: bucket acme-tfstate, key staging/terraform.tfstate, region eu-central-1, dynamodb_table = "acme-tfstate-locks", encrypt = true.
Tracked and unmodified: envs/staging/terraform.tfstate - version 4, terraform_version 1.9.5, serial 41, lineage a3f1, "resources":[]. Identical to the copy at HEAD; git show HEAD:envs/staging/terraform.tfstate is currently the only second copy that exists anywhere.
No .gitignore in the repo. No envs/prod directory - prod's state was never located this session.
Last verified: git status --porcelain - one line, ?? envs/staging/backend.tf.
Last verified: find C:/Users/Nihil/AppData/Local/Temp -name "*tfstate*" - exactly one hit, the file in the repo.
Tree disagrees with session: the session said it copied the state file to a temp location before touching anything. The search above finds no such copy. Git history is the only backup.
Tree disagrees with session: the ticket premise is that the committed state contains AWS resource ids. The file's resources array is empty, while envs/staging/main.tf declares aws_s3_bucket.assets and aws_db_instance.primary. Either this state is not the live state, or the ids the audit cares about are somewhere else.
Settled
- S3 + DynamoDB backend, not Terraform Cloud - because Terraform Cloud is priced per user and finance has frozen new SaaS seats until at least January; Tomek said so on Monday when the Datadog request was refused the same way - session decision.
- Backend parameters as written: bucket
acme-tfstate, keystaging/terraform.tfstate, regioneu-central-1, lock tableacme-tfstate-locks,encrypt = true-envs/staging/backend.tf:1-9. - Backend declared per environment rather than extracted into a shared module - because extraction only pays off once a third environment exists -
modules/backend/README.md. - Bucket versioning turned on for
acme-tfstatethrough the AWS console, accepted as an out-of-band change because the bucket is not managed by this repo, so there is nothing to drift against - session decision, see Believed. - Staging migrates first and alone; prod is a separate change - session decision.
Open
- The migration has not run.
backend.tfexists but no terraform command has ever consumed it, so the local file is still the real state - blocks: all of INFRA-88 - settled by:terraform init -migrate-stateinenvs/stagingsucceeding, and the state appearing ats3://acme-tfstate/staging/terraform.tfstate. envs/staging/terraform.tfstaterecords no resources at serial 41 whilemain.tfdeclares two. Aterraform planafter migration will propose creating an S3 bucket and an RDS instance that probably already exist - blocks: anyapplyon staging - settled by:terraform planinenvs/stagingand reading what it proposes before anyone applies.- Is prod in scope for INFRA-88, or does it get its own ticket?
- blocks: knowing whether this ticket closes before the audit - settled by: asking Priya, who owns INFRA-88.
- Key layout for prod: one bucket with
prod/andstaging/prefixes, or separate buckets with separate policies - blocks: the prod migration - settled by: a decision from Priya alongside the scope question. - Whether
envs/staging/terraform.tfstategetsgit rm'd and gitignored once the migration lands. If it stays tracked, the audit finding survives the migration - blocks: the audit objective the ticket exists for - settled by: owner decision plus a commit; there is no.gitignoreto add to yet.
Tried and abandoned
- Terraform Cloud instead of a self-run bucket - died on: per-seat pricing against a finance freeze on new SaaS seats - retry only if: the freeze lifts, earliest January, and then as a new ticket rather than INFRA-88.
- First
terraform init -migrate-staterun - died on:Error: Failed to configure the backend "s3": Bucket versioning is not enabled for "acme-tfstate"- retry only if: versioning is confirmed on the bucket, which is exactly the retry now pending.
Believed, not verified
- Versioning is now enabled on
acme-tfstate; the console change produced no captured output - confirm with:aws s3api get-bucket-versioning --bucket acme-tfstate. - The lock table
acme-tfstate-locksexists. Nothing in this session created or checked it - confirm with:aws dynamodb describe-table --table-name acme-tfstate-locks. acme-tfstateis ineu-central-1, asbackend.tfasserts - confirm with:aws s3api get-bucket-location --bucket acme-tfstate.- Serial 41 is the live staging state. No terraform command ran successfully in this session - confirm with:
terraform planinenvs/staging.
Not in scope
- The prod environment. It has no directory in this repo and was deliberately left for a separate change.
- Extracting
modules/backendinto a real module. Left as a placeholder README on purpose. - The resources in
envs/staging/main.tf. No infrastructure was created, changed or destroyed this session; the only AWS-side change was enabling bucket versioning. README.mdstill says state is committed by hand. True today, and left to be corrected by whoever lands the migration.
session-handoff-brief.pdf
PDF · document
Example file from a real run - the skill writes it into your workspace.
Subscription · always the latest version
- Secure checkout via Stripe
- Cancel anytime
- Delivered live via MCP
About this skill
Long agent sessions end the same way: the context window fills, and the next session starts by rediscovering what the last one already knew. This skill turns the session you are in into a brief another agent can resume from cold - one next action, the real state on disk, what is settled, what is still open, and what was already tried. Every dead end carries the reason it died, so nobody retries it. Anything nobody actually ran is marked believed, not verified.
What it does
Turns the agent session you are in into a handoff document a different session can resume from without having seen any of it - one next action with the file to open, the real state on disk, what is settled, what is still open, and what was tried and abandoned with the reason it died. It reads the working tree as well as the conversation, so a claim the session believes and the repo contradicts is reported as a disagreement rather than quietly resolved.
Try it first
At the end of a session where you actually got somewhere, say "I'm running out of context - write a handoff for the next session". You should get back a Handoff with its boundary on the first line, a single Resume here action naming a file, a State on disk block with the branch and the uncommitted files, then Settled, Open, Tried and abandoned, and Believed not verified - each entry anchored to a path, a command, or a decision it can quote.
What it needs
An agent that can read and search the working tree - Read, Grep, Glob and Bash. No API keys, no MCP servers, no network access. Git is not required, but without it the on-disk half is unavailable and the skill says so in the first line instead of guessing.
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Agents & Orchestration.
It cannot recover what the session no longer holds. If the context was already compacted, or you ask for a handoff before any real work happened, the brief is thin - and it will say it is thin rather than pad it. It does not commit, stash or clean anything, so a half-finished edit is still yours to deal with; the brief only points at it. It also does not write the file until you ask, which means one extra turn if you wanted HANDOFF.md straight away. And the on-disk half is only as good as the repo: work spread across several repos, or living in untracked scratch directories, gets recorded from the conversation alone and lands under believed, not verified.
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
Free to use
No account required to browse
Trust & safety
Security scan
Security scan passed 1 month ago