Engineering skills with receipts.

Canonical contributor-side intake gate. It qualifies a bounded shortlist, keeps the final choice human, optionally uses xref as graph evidence, and seeds the Issue Contract without owning reproduction or implementation.

Used to capture real public maintenance work and backfills as evidence ledgers. Promoted to the stable distribution surface; it has not completed a baseline comparison.

Pre-review gate harvested from maintainer review rounds across agent-browser and portless cases. Blind replication round 002 compared the method against an answer key and exposed both catches and verification losses.

Orchestrates Shaping before implementation: freezes the evidence-backed contract, runs isolated Shaping passes, probes weak assumptions, feeds evidence back into fit checks, and gates the surviving shape before detailing. Promoted to stable on Hunter's direct call; formal baseline comparison remains pending.

Niche triage for bugs that require unavailable hardware or platforms, using structural or visual signatures on available machines. Applied as written on agent-browser #1461 and needed no revision. Baseline eval round still pending.

Append-only decision trail written to the canonical Foundry during implementation so review can audit choices, not only the diff. Redesigned after the lifecycle audit and awaiting one real dogfood run.

Browser-openable before/after artifacts for visual behavior, benchmark deltas, and feature tours. Registered after repeated personal use across wterm work; no baseline comparison has been completed.

End-of-cycle handoff derived from twelve hand-written handoffs across vcut, petdex, portless, wterm, gateway-spend, and agent-browser. Promoted on Hunter's direct call after heavy personal dogfooding; a dedicated repository case is still owed.

Measurement-first performance workflow. Frozen outside the default issue workflow and invoked only for an evidenced performance claim.

Read-only repository quality baseline. Ran on portless #369 and surfaced the untested write paths that led directly to the fix. Baseline comparison remains pending.

Failure-path audit. Frozen outside the default issue workflow and invoked only when timeout, cancellation, retry, cleanup, overload, or concurrency risk triggers it.

Bounded behavior-preserving reduction workflow derived from a real feature simplification below a PR line budget. Deterministic, Test Strength, Resilience, and external review gates passed; baseline comparison remains pending.

Test effectiveness workflow using falsification, mutation, properties, real boundaries, and artifact checks. Applied inside real Review Gate runs on portless #366 and #374. Baseline comparison remains pending.

Read-only admission assessment for one already-selected issue, pull request, or manual request. It classifies intent, recommends the minimum workflow, and stops for human confirmation before mutation authority is granted.

Revalidates handoff and portfolio claims against live GitHub, local Git, releases, deployments, and package artifacts, then classifies drift and stalled work into a current read-only queue. Applied twice on real Vercel Labs portfolio work; no baseline comparison yet.

Snapshots the reference graph of a GitHub pull request or issue to surface orphans, competing work, structural links, and file-overlap risks. Registered after repeated real use; no baseline comparison yet.