wrap-up

Name: wrap-up
Author: anthropics

832stars159forks

How to install wrap-up?

anthropics/launch-your-agent/wrap-up

$npx -y skills add anthropics/launch-your-agent --skill wrap-up

Installs into the current project.

Summary

Close out (or revisit) a Claude Managed Agent build — refresh the overview page, recap every primitive the founder now owns, show the run log and live status, suggest 1–2 tailored next upgrades, and sweep hygiene (sessions archived, key only in .env, no literal dates in deploymen

Files · 1

View on GitHub

SKILL.md

1	<!-- Copyright 2026 Anthropic PBC -->
2	<!-- SPDX-License-Identifier: Apache-2.0 -->
3
4	# Wrap-up — celebrate it, show what they built, point at what's next
5
6	You are closing out (or checking in on) a founder's Claude Managed Agent. The tone is celebratory: they just shipped a managed agent that runs without them — say so, warmly, in one or two sentences ("🎉 you've shipped a managed agent — here's what you built"), then show them what they own and the one or two upgrades worth doing next. End on the overview page, not a wall of text. This is a moment, not a checklist read-out.
7
8	Follow the same voice rules as `/launch-your-agent`: warm, compact, tables for anything enumerable, primitives called by their real names (Outcome, Session, Deployment, Vault, Memory store) with a plain gloss on first use, the shared emoji shorthand (🤖 agent · 📦 environment · 🎯 outcome · ▶️ session · 🗓️ deployment · 🔌 connector · 🔐 vault · 🧠 memory store · 🧪 evals), no unverified timings. No cost notes — don't volunteer per-run or per-month spend figures; answer cost questions only if the founder asks.
9
10	## 1. Read the state
11
12	- Read `./my-agent/`: `build-sheet.json`, `IDS.env`, `NEXT-DIRECTIONS.md`, `outcome.md`, `evals/`, the existing `agent-overview.html`. (If there is no `my-agent/` folder here, say so and stop — point them at `/launch-your-agent`.)
13	- If `my-agent/.env` holds a key, `source` it and refresh live state via the API (shapes: `../launch-your-agent/references/cma-api.md`): each session's `status` + `outcome_evaluations[]`, deployment `status` + `schedule.upcoming_runs_at`, latest `usage`. Parse with python (`strict=False`), never jq.
14	- No key → continue describe-only and say in one sentence that live status needs the key in `.env`.
15	- A session still `running` → say so and ask (AskUserQuestion): wait for it, or wrap with what's done.
16	- Nothing launched yet → wrap the plan only; everything below still applies but the status is "○ Planned".
17
18	## 2. Produce the wrap-up (in this order)
19
20	1. Congratulate them. One or two warm sentences: they shipped a managed agent — name it, say what it now does on its own. 🎉
21	2. Overview page. Refresh `agent-overview.html` (template: `../launch-your-agent/references/overview-template.html`; styling in the sibling `overview.css`, linked, never inlined). The page already exists from the build — Edit only the slots that changed (status pill, top-bar IDs/stat cells, run-log `<tbody>` rows, eval verdicts, schedule with next run times, Console links for the key's workspace, the v1/v2 next directions) rather than re-Writing the whole file. If either `agent-overview.html` or `my-agent/overview.css` is missing, `cp` the CSS from the launch-your-agent skill's `references/` first, then Write the HTML once. Open it in their browser — the page is the closing artifact; the chat below just points at it.
22	3. "Here's what you built" — the primitives recap table, one row per CMA primitive that now exists: emoji · what it is (one plain sentence) · what it's set to · live ID · which card on the page shows it. Final muted row(s) for primitives deliberately not used → "see NEXT-DIRECTIONS". Follow with the run log table (run · rubric version · verdict · one-line note).
23	4. "Here's what's next" — 1–2 extensions picked from the v1/v2 plan that matter most for this use case, pitched concretely: what it does for them, what it takes, how small the change is. The rest of the plan stays written down. Standard candidates to weigh alongside whatever the plan already holds: delivery via a connector (Slack/email, `always_ask`-gated), a memory store if runs repeat themselves, wiring a mocked connector for real, and — when it suits how they'll actually use the agent — a generated interface: a page or small app Claude Code builds for them in a follow-up session, shaped to the need (graphical output for results that want charts, a results viewer that pulls outputs and verdicts via the Files/Sessions API, or a simple way to interact with the agent — kick off a run, steer it, answer its confirmations).
24	5. Hygiene sweep — quiet by default. Do the sweep (archive finished sessions; check the deployment's `initial_events` for literal dates; check the key only ever lived in `.env`; save a passed run as `evals/case-01/expected.md` if there's no golden case yet) a

People who liked this also viewed

6 picks

Type

wrap-up

832stars159forks

View on GitHub

How to install wrap-up?

anthropics/launch-your-agent/wrap-up

$npx -y skills add anthropics/launch-your-agent --skill wrap-up

Installs into the current project.

Summary

Files · 1

View on GitHub

SKILL.md

1	<!-- Copyright 2026 Anthropic PBC -->
2	<!-- SPDX-License-Identifier: Apache-2.0 -->
3
4	# Wrap-up — celebrate it, show what they built, point at what's next
5
6	You are closing out (or checking in on) a founder's Claude Managed Agent. The tone is celebratory: they just shipped a managed agent that runs without them — say so, warmly, in one or two sentences ("🎉 you've shipped a managed agent — here's what you built"), then show them what they own and the one or two upgrades worth doing next. End on the overview page, not a wall of text. This is a moment, not a checklist read-out.
7
8	Follow the same voice rules as `/launch-your-agent`: warm, compact, tables for anything enumerable, primitives called by their real names (Outcome, Session, Deployment, Vault, Memory store) with a plain gloss on first use, the shared emoji shorthand (🤖 agent · 📦 environment · 🎯 outcome · ▶️ session · 🗓️ deployment · 🔌 connector · 🔐 vault · 🧠 memory store · 🧪 evals), no unverified timings. No cost notes — don't volunteer per-run or per-month spend figures; answer cost questions only if the founder asks.
9
10	## 1. Read the state
11
12	- Read `./my-agent/`: `build-sheet.json`, `IDS.env`, `NEXT-DIRECTIONS.md`, `outcome.md`, `evals/`, the existing `agent-overview.html`. (If there is no `my-agent/` folder here, say so and stop — point them at `/launch-your-agent`.)
13	- If `my-agent/.env` holds a key, `source` it and refresh live state via the API (shapes: `../launch-your-agent/references/cma-api.md`): each session's `status` + `outcome_evaluations[]`, deployment `status` + `schedule.upcoming_runs_at`, latest `usage`. Parse with python (`strict=False`), never jq.
14	- No key → continue describe-only and say in one sentence that live status needs the key in `.env`.
15	- A session still `running` → say so and ask (AskUserQuestion): wait for it, or wrap with what's done.
16	- Nothing launched yet → wrap the plan only; everything below still applies but the status is "○ Planned".
17
18	## 2. Produce the wrap-up (in this order)
19
20	1. Congratulate them. One or two warm sentences: they shipped a managed agent — name it, say what it now does on its own. 🎉
21	2. Overview page. Refresh `agent-overview.html` (template: `../launch-your-agent/references/overview-template.html`; styling in the sibling `overview.css`, linked, never inlined). The page already exists from the build — Edit only the slots that changed (status pill, top-bar IDs/stat cells, run-log `<tbody>` rows, eval verdicts, schedule with next run times, Console links for the key's workspace, the v1/v2 next directions) rather than re-Writing the whole file. If either `agent-overview.html` or `my-agent/overview.css` is missing, `cp` the CSS from the launch-your-agent skill's `references/` first, then Write the HTML once. Open it in their browser — the page is the closing artifact; the chat below just points at it.
22	3. "Here's what you built" — the primitives recap table, one row per CMA primitive that now exists: emoji · what it is (one plain sentence) · what it's set to · live ID · which card on the page shows it. Final muted row(s) for primitives deliberately not used → "see NEXT-DIRECTIONS". Follow with the run log table (run · rubric version · verdict · one-line note).
23	4. "Here's what's next" — 1–2 extensions picked from the v1/v2 plan that matter most for this use case, pitched concretely: what it does for them, what it takes, how small the change is. The rest of the plan stays written down. Standard candidates to weigh alongside whatever the plan already holds: delivery via a connector (Slack/email, `always_ask`-gated), a memory store if runs repeat themselves, wiring a mocked connector for real, and — when it suits how they'll actually use the agent — a generated interface: a page or small app Claude Code builds for them in a follow-up session, shaped to the need (graphical output for results that want charts, a results viewer that pulls outputs and verdicts via the Files/Sessions API, or a simple way to interact with the agent — kick off a run, steer it, answer its confirmations).
24	5. Hygiene sweep — quiet by default. Do the sweep (archive finished sessions; check the deployment's `initial_events` for literal dates; check the key only ever lived in `.env`; save a passed run as `evals/case-01/expected.md` if there's no golden case yet) a

People who liked this also viewed

6 picks

Type