Release intelligence, run as a CLI

Landmark computes what should ship next — and refuses to write a release note it can't prove.

Landmark turns a repo's conventional commits into a version decision, a technical changelog, and public release notes — with a machine-readable evidence packet behind every claim. Every section it writes has to trace back to a real commit; if it can't, it refuses to write the section rather than guess.

Product boundary is the Rust CLI — no framework lock-in, no required GitHub Action.

Why it matters

job Decide the next semantic version from real commit history and API-diff evidence, not guesswork. proof docs/releases/v1.28.0.md — the actual notes this run wrote, sourced from 8 real commits in v1.27.0..HEAD.

job Refuse to fabricate a release-note section it can't ground in a real commit. proof Shipped in crates/landmark/src/claim_grounding.rs — the honest-refusal gate is the point, not a caveat.

job Plan final-mile artifacts — docs updates, blog drafts, demo scripts — as typed, reviewable producer contracts. proof Real run output: Landmark-owned artifacts ship not-required; producer artifacts sit pending for a human, always.

job Run the same runtime from a laptop, generic CI, or an agent — the CLI is the product, the Action is optional. proof

cargo run --locked -p landmark -- run --provider local
                      --repo-root .

Real terminal output

Landmark has no UI to screenshot, so this gallery is real terminal and file captures from a live run against this repo's own history — not mockups. Click a frame to inspect it.

Release notes

Release notes are public and user-facing, sourced from Landmark's own docs/releases/releases.json export seam — the same artifact a marketing site can render without loading Landmark or re-reading git history.