List the most recently benchmarked commits from master_commits, newest first. Note: the full sha returned here must be truncated to its first 10 hex characters before using it as the sha filter in list_summaries.

List compiler benchmark summary rows for a given commit short-sha (first 10 hex chars). Each row's summary field is a JSON blob with detailed compile-time and cycle-count metrics. Optionally filter by test name (exact) or a substring of it, and by cloud/context. Supports pagination.

Fetch one summaries row by its UUID id, including the full summary JSON metrics blob.

Build a dashboard.furiosa.dev/ir-text-viewer link for a given commit + context + test, so it can be opened directly in a browser instead of running anything locally. Resolves the short-sha to the full commit sha required by the viewer URL via master_commits.

DRAFT / PROPOSED contract only -- dashboard.furiosa.dev does not read this recipe param yet. This exists so an AI agent can translate a loose display request ('bridge에 하이라이트 해줘', 'IO 연산은 파란색으로', 'latency 큰 operator 강조해줘') into a small, fixed set of composable view directives instead of re-deriving a full custom rendering from raw IR data on every request. The tool only validates and encodes the recipe as a query param; a deterministic renderer in the dashboard frontend (separate codebase) still needs to be built to read recipe and actually draw it. Until that exists, treat the returned URL's recipe param as inert and the response's recipe field as the artifact to hand to whoever implements that renderer.

Build a compare link on ir-viewer.furiosa.dev (a separate viewer from dashboard.furiosa.dev's ir-text-viewer) for a base vs current commit. Accepts short-shas directly, no full-sha lookup needed.

Build a tactic-debug compare link for a base vs current commit. Accepts short-shas directly.

Build a tc-viewer compare link for a base vs current commit. Unlike the other two compare viewers, tc-viewer takes a single (not comma-paired) context/test and separate sha/baseSha params.

Diff all matching benchmark rows (matched by test+context+cloud) between a base and current commit short-sha. Reports added/removed tests, pass/fail status changes, the top N regressions/improvements for both execution-style tests (status + average_actual_cycle) and compile-style tests (rlir total_cycle + total compile_time), plus a generic fieldChanges section that flattens and ranks every numeric field found anywhere in the summary JSON blob (memory, tactic costs, dram usage, per-phase cpu time, etc.) -- use fieldChanges as a first-pass scan for what's worth attention before drilling into a specific metric. Optionally scope the comparison with test/testContains/context/cloud filters, same as list_summaries.