Launch Smoke Test

A deliberately tiny launch skill whose only purpose is to prove the tessl launch skill --cloud pipeline works end to end. It touches both external systems a real launch skill depends on -- Linear (read) and GitHub (write) -- with the smallest possible footprint, so a green run means the sandbox got its credentials, can talk to both services, and can open a PR.

It does three things, and only these three:

1. Reads one Linear issue through the Linear API (read-only -- proves LINEAR_TOKEN is present and valid).
2. Appends a single run-marker line to a log file (the small change).
3. Opens a pull request containing that one-line change (proves gh / GITHUB_TOKEN and the publish path work).

It makes no Linear writes and produces no other repo changes. If you are looking for a skill that does real work, this is not it -- this is the smoke test you run to confirm the pipeline is healthy before trusting it with real work.

Inputs

The launcher supplies these via .cloud-launch/inputs.json. All are optional; the defaults make the skill runnable with no inputs at all.

The environment must carry:

• LINEAR_TOKEN -- a Linear API token authorised to read issues. Used only for the single read below.
• GitHub credentials -- GH_TOKEN / GITHUB_TOKEN so the gh CLI can push the branch and open the PR.

If either is absent, stop and report the missing credential rather than guessing -- a missing credential is exactly the kind of pipeline break this skill exists to surface.

Procedure

You are running in a clone of the target repo, checked out on its base branch, at the repository root. Git already has a credential helper and a bot commit identity configured, so git push and gh pr create authenticate without any setup -- do not reconfigure auth, and never print or commit the tokens.

1. Read the Linear issue

Fetch the issue named by {{ISSUE_IDENTIFIER}}. Linear's issue(id:) query accepts the human identifier (e.g. AE-669) directly:

curl -s https://api.linear.app/graphql \
  -H "Authorization: $LINEAR_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"query":"query($id: String!) { issue(id: $id) { identifier title url state { name } } }","variables":{"id":"{{ISSUE_IDENTIFIER}}"}}'

Keep the identifier and title from the response. If the response carries a GraphQL errors array or no issue, stop and report it -- the read is the Linear half of the smoke test, so a failure here is a real signal, not something to work around.

2. Append the run marker

Build a one-line Markdown bullet recording this run, then append it to {{LOG_FILE}} (create the file with a # Launch smoke-test log heading first if it does not yet exist):

- <UTC timestamp> — cloud-launch smoke test — read <identifier>: "<title>"

Use date -u +%Y-%m-%dT%H:%M:%SZ for the timestamp. Embedding the fetched issue title is what ties the two halves together: the line in the PR is direct evidence the Linear read succeeded.

3. Open the pull request

Branch, commit just the log file, push, and open the PR:

BRANCH="{{BRANCH_PREFIX}}/$(date -u +%Y%m%d-%H%M%S)"
git checkout -b "$BRANCH"
git add "{{LOG_FILE}}"
git commit -m "chore(kikimora): launch smoke-test run marker ({{ISSUE_IDENTIFIER}})"
git push -u origin "$BRANCH"
gh pr create \
  --title "chore(kikimora): launch smoke-test run marker ({{ISSUE_IDENTIFIER}})" \
  --body "Automated smoke test of the \`tessl launch skill --cloud\` pipeline. Read \`{{ISSUE_IDENTIFIER}}\` from Linear and appended a run marker to \`{{LOG_FILE}}\`. This PR's existence proves the launch reached Linear and GitHub and opened a PR end to end."

gh pr create targets the repository's default branch, which is the base the clone was checked out on -- that is what you want. Capture the PR URL it prints.

Then stop. Do not write launch-result.json -- recording the outcome is the recipe's job, not the skill's; it inspects the branch and PR you left behind. The recipe records that outcome under the fixed launch-result envelope's output property, shaped by this skill's declared schemas.output (see Typed output below) -- the PR you just opened, keyed by pr_url, branch, pr_number, and issue_identifier.

Notes

• Idempotency / concurrency. Each run appends a new, uniquely-timestamped line and opens its own branch, so re-running is always safe. Two runs in flight at once against the same {{LOG_FILE}} can conflict at merge time -- expected for an append-to-shared-file smoke test; land them one at a time or point concurrent runs at different {{LOG_FILE}} paths.
• Why a real PR and not a dry run. Opening an actual PR is the point: it is the only thing that exercises push + gh pr create + the bot identity. The diff is a single harmless log line, trivial to close or revert.
• Typed output. The frontmatter declares a schemas.output (a local $ref to ./schemas/output.schema.json) describing the PR this skill opens. On a cloud launch, enforce-skill-schemas composes that schema into the fixed launch-result envelope's output property (making output required) and the recipe validates the recorded result against it, so a launch of this skill exercises the typed-output path end to end -- the smoke test's second job. Declaring the contract is not "doing work": the skill still opens exactly the same one-line-diff PR and still writes no launch-result.json itself.

What this skill deliberately omits

• No Linear writes. It only reads. Proving the write path is the dark-factory-labeller's job, not this skill's -- keeping this one read-only makes a failed run unambiguous (a write failure can't be confused for a read failure).
• No launch-result.json. The recipe's launch task records the outcome by inspecting the branch and PR; a skill that wrote it would be doing the recipe's job.
• No real work. The appended line carries no meaning beyond "a launch ran here." Do not extend this skill to do useful work -- create a purpose-built skill for that and keep this one minimal so it stays a fast, unambiguous pipeline check. (Declaring schemas.output is not work: it types the PR the skill already opens so a launch also proves the typed-output path, without changing what the skill does.)