| name | landmark-dogfood |
| description | Dogfood Landmark fleet adoption and release-note automation across GitHub repositories. Use when asked to integrate Landmark into personal or organization repos, run fleet scan/plan/open-prs, evaluate adoption friction, capture release-pipeline evidence, or turn dogfood findings into Landmark fixes/backlog. |
Landmark Dogfood
Use Landmark against real repositories before claiming adoption quality. The output is not just a plan: it is an evidence packet, concrete downstream PRs where safe, and upstream fixes or backlog where the dogfood loop exposes product gaps.
Workflow
- Start from the local Landmark checkout and read
AGENTS.md.
- Create an evidence directory under
.landmark/dogfood/<date-or-run>/.
- Build/run the Rust binary locally with
cargo run --locked -- ... or target/debug/landmark. The action itself bootstrap-downloads a published release binary; there is no checked-in binary to run locally.
- Run a read-only fleet scan first:
target/debug/landmark fleet scan \
--owner phrazzld \
--owner misty-step \
--active-only \
--output .landmark/dogfood/<run>/active-scan.json
- Run
--deep-checks before opening or recommending PRs; default scans intentionally mark secret metadata unavailable.
- Generate
fleet plan, then fleet open-prs --dry-run; inspect the rendered files before mutating downstream repos.
- Use the plan's
repository_kind, release_surface, integration_mode, and integration_rationale fields as the first-pass rollout map. Separate applications, libraries, infrastructure, archived repos, experiments, non-release repos, no-release-tool repos, and already-adopted repos. Do not treat every active repo as an app.
- For each candidate, inspect existing workflows, releases, tags,
AGENTS.md, and release notes. Existing Landmark workflows should become manifest/upgrade work, not duplicate workflow installation.
- Apply downstream integration only when the generated diff is repo-fit and required secrets are present or intentionally provisioned. Use
fleet open-prs --confirm-remote --max-prs 1 only after the dry-run receipt is inspected; never provision secrets for local, generic-ci, manifest-only, backfill-first, or skipped non-release modes just to make a GitHub plan look ready.
- Roll out one repository at a time. After a downstream PR merges, monitor the release run or local/generic CI artifact path named in the receipt before continuing the fleet.
- Record friction immediately. Use references/finding-taxonomy.md for categories and severity.
- Fix Landmark itself when the product made the dogfood run unsafe, confusing, or wrong. Add replay coverage for each fixed failure mode.
Evidence
Every run should leave:
active-scan.json and, when used, deep-active-scan.json
plan/plan.json and plan/README.md
- dry-run PR artifacts under
plan/prs/
plan/prs/open-prs.json receipts with branch, commit message, rollback, disposition, evidence directory, monitoring status, and APPLY.md for confirmed one-repo rollout packets
- a short
report.md listing adopted repos, blocked repos, friction, fixes made, and next rollout steps
- exact commands and hosted PR/check URLs for downstream integrations
Safety
- Prefer env-based token discovery. Avoid passing secrets as CLI arguments, especially through
cargo run, because Cargo echoes arguments.
- Never print secret values. Search dogfood artifacts for token-like strings before committing.
- Do not set or overwrite repository secrets in bulk until the target set is narrowed to real applications and the token/value source is explicit.
- Do not open duplicate release workflows for repos that already invoke
misty-step/landmark.
- Do not add GitHub workflows for
local, generic-ci, manifest-only, or backfill-first plans unless repo-specific review deliberately overrides the generated receipt.
- Treat generated
open-prs output as a proposal, not authority. The lead agent owns repo-fit.