# rea-ci-tests-audit Audit REA CI and test suites for correctness, fixture drift, false passes, and measured performance; use for suite or workflow investigations and efficiency changes, rather than ordinary feature testing or a single test failure. - **Kind:** skill - **Source:** https://github.com/morluto/rea - **Page:** https://forefy.com/skills/6f363161-182e-4d13-95aa-5a69cf43abee - **API (JSON + files):** https://forefy.com/api/asr/6f363161-182e-4d13-95aa-5a69cf43abee ## Install Ask your agent: `Install the rea-ci-tests-audit skill securely from: https://forefy.com/skills/6f363161-182e-4d13-95aa-5a69cf43abee globally to ~/.claude/skills` Or run the pinned install script: `curl -fsSL https://forefy.com/api/asr/6f363161-182e-4d13-95aa-5a69cf43abee/install | bash` (`?location=project` for the current project, `&harness=agents` for `.agents/`, `&shell=powershell` with `irm ... | iex`). --- ## SKILL.md --- name: rea-ci-tests-audit description: Audit REA CI and test suites for correctness, fixture drift, false passes, and measured performance; use for suite or workflow investigations and efficiency changes, rather than ordinary feature testing or a single test failure. --- # REA CI and tests audit Establish which checks still catch real regressions and which costs can be removed without losing that protection. The outcome may be a verified fix, an evidence-backed recommendation, or an explicit decision to stop. Follow the checkout's `AGENTS.md`, [testing guide](../../../docs/testing.md), and [contribution rules](../../../CONTRIBUTING.md). Use their current commands and pinned toolchain, rather than commands remembered from an earlier audit. Preserve the requested scope: investigation alone does not authorize edits; this skill does not grant permission to publish issues, push, or merge. Existing user authorization still applies. Keep audit notes and raw results outside the tracked tree, normally in a task-specific `/tmp` directory. ## Establish the executed baseline Fetch when the user asks for latest main. Pin the baseline SHA and inspected workflow revision; use an isolated worktree when changing or comparing code. Do not modify another task's checkout. Record dirty changes and the exact command, toolchain, host, selected tests, instrumentation, and prerequisites. An old run is evidence for its executed revision, even if main advances. Trace workflow triggers, changed-path selection, job dependencies, reusable actions, package scripts, test discovery, artifact producers, and report gates through to the commands that actually execute. Include scripts, generated data, runtime registration, and subprocess boundaries missing from import graphs. Check positive and negative selection paths: required files must be discovered once per intended test configuration; deliberate platform/configuration repeats are separate coverage. A missing explicit test or missing required report must not turn green. Optional lanes may skip; preserve their stated prerequisites. Use existing timing and failure artifacts to select suspicious families. Run relevant checks to establish a baseline; reserve a full local suite for broad CI diagnosis or uncertainty that focused checks cannot resolve. Do not rebuild or rerun an unchanged complete suite after every documentation edit. ## Prove the failure or protection For each candidate, state the expected observable behavior and its owner, then construct the smallest reproduction against production behavior. Follow the producer and affected consumers before deciding where to fix it. - For drift, compare fixtures with current admission requirements, protocol replies, named contracts, tool versions, and actual public entrypoints. A plausible-looking mock can fail before the behavior its test claims to cover. Search the whole affected fixture family, including other version strings. Preserve intentionally malformed fixtures and test their intended rejection. - For a suspicious green test, perturb the relevant behavior in an isolated copy or inject a controlled failure. Verify that the existing assertion or gate detects it. A test may merely repeat its mock, import abandoned code, or hide a subprocess/checker failure. Restore mutations after the experiment. - For shared setup or pruning, map distinct assertions to their retained scenarios. Preserve malformed inputs, cancellation, cleanup failure, permissions, capacity, identity, and meaningful concurrency interactions. A faster run that skips a case or weakens its oracle is not an improvement. - For provider coverage, separate fake-producer, boundary, and actual-engine evidence. Inspect named skips and prerequisites; fixtures do not establish real Ghidra, LLDB, browser, or other provider behavior. Classify findings as reproduced defects, supported costs, hypotheses, or unavailable evidence. Record the baseline command, observed output/exit status, expected result, affected scope, and competing explanation. A failure on main may be pre-existing; that does not by itself make it irrelevant to a CI audit. Attribute PR failures using the executed revision and a clean baseline, not just their resemblance to a previous failure. Retry only to test a specific transient explanation; retain the original failure and stop when repetition adds no evidence. A passing retry does not establish a root cause or a fix. ## Assess efficiency without losing the gate Read [measurement and CI checks](references/measurement-and-ci.md) when investigating slowness, caching, runner waits, shared artifacts, or log noise. Compare equivalent completed work before claiming a speedup. Distinguish setup/build time, test execution, coverage instrumentation, dependency waits, and runner scheduling; state which costs remain unmeasured. Prefer removing proven duplicate preparation, reusing immutable setup, and reducing repeated subprocess work before shrinking correctness coverage. Preserve independent sessions where lifecycle or concurrency is the oracle. Keep live tests fresh; a cached previous pass cannot validate current host, installed provider contents, permissions, or external state. ## Fix, verify, and hand off within scope If implementation is authorized, make the smallest structural change at the owning boundary and add a regression that fails on the baseline for the right reason. Retest affected consumers and meaningful negative paths; verify cache invalidation or discovery when those mechanisms change. Read the installed Turbo documentation before changing its configuration or commands. Run the repository's applicable checks and inspect complete current-head CI results before an authorized merge. Preserve required coverage and discovery gates. Do not suppress a failure, add blanket retries, raise timeouts, or delete distinct assertions solely to make CI green. Changes to these controls need their own demonstrated reason and retained failure detection. Report each confirmed issue with exact revision, reproduction, consequence, proposed correction, and validation limits. Report performance with workload and measurements, not just a percentage. Use English issue/PR records when requested; follow the user's assignment, commit identity, review, and merge preferences without assuming them for other users. Open ready-for-review PRs only when that lifecycle is authorized. Resolve necessary review findings and check final main against the tested tree after merging. Stop when the authorized work is verified, remaining benefits are unproven, or further changes would trade correctness for speculative speed. List genuine uncertainties without turning them into additional optimization rounds. ## agents ``` ``` ## agents/openai.yaml ```yaml interface: display_name: "REA CI and tests audit" short_description: "Audit CI/test correctness and measured performance" default_prompt: "Use $rea-ci-tests-audit to investigate CI and test correctness, drift, and performance before proposing changes." ``` ## references ``` ``` ## references/measurement-and-ci.md # Measurement and CI checks ## Equivalent completed work Record the source/workflow SHA, command, test identities and counts, skips, assertions or output contract, exit status, coverage scope and thresholds, toolchain/provider versions, host resources, and prepared artifacts. Inspect both the child result and its report: a profiling wrapper may exit successfully after the measured command failed. Focused tests can pass while global coverage thresholds fail; describe both rather than calling the command green. Measure wall time, user/system CPU, and peak process-family memory for relevant completed tasks. Parent-only RSS misses workers and their children. On Linux, summed RSS can double-count shared pages; report RSS/PSS semantics, sampling interval, missing samples, and observed cleanup limits. Use equivalent host tools elsewhere. State unknown costs, including agent tokens; fewer log bytes alone prove neither fewer tokens nor faster tests. Separate cold execution, warm build/file caches, and warm runtime caches. Use repeated or alternating runs when variance affects the decision, with enough observations to distinguish the proposed benefit from noise. Keep workload, instrumentation, and host conditions comparable; record interference. A run with fewer tests, early rejection, lost coverage, or a different runner does not prove suite acceleration. CI-wide claims need end-to-end measurements, not an extrapolation from one improved family. ## Workflow, caches, and shared artifacts Trace each reused result from producer to consumers and identify its lifetime. Check key/input coverage against the actual file inventory and result-affecting configuration, environment, generated data, tool versions, and host state. Exercise an affected input change to prove invalidation; exercise an irrelevant change to support reuse where useful. Cache deterministic preparation and static analysis only when their inputs fully describe the result. Verify downloaded artifacts match the expected revision, platform, runtime, and schema. Check restore misses, failed/partial producers, stale reports, removed output files, and consumers after producer failures. Shared writers need output ownership or locks; tests need isolated state, not an unnecessary global lock. Worker counts do not bound children spawned within each test. Preserve intentionally concurrent scenarios while bounding unrelated batches. Check matrix expansion and all job pages before saying CI passed. Inspect required gates after failed, cancelled, and skipped dependencies. `always()` alone does not preserve failures; inspect the aggregation logic. Test/report uploads should retain available diagnostics after failure, while absent or malformed required results must fail closed. Do not demand an optional report from a lane intentionally excluded by the selector. ## Runner waits and console noise Use per-job and per-step timestamps where available. Workflow creation to job start includes dependencies and other delays; it is not necessarily runner queue time. Identify when the job became eligible before attributing waits to runner capacity. If eligibility/queue timestamps are unavailable, report the combined delay and unknown split. Compare equivalent lanes and runner types across enough runs; a single busy runner is not a workflow regression. For console reductions, retain checker exit codes, actionable errors, useful summary counts, and complete machine-readable artifacts. Verify authored code remains in the actual checker inventory when excluding generated outputs. Introduce an authored error as a control; it must still fail. An ignore pattern can hide real source as well as noise. Show that cached reports are restored, stale successful reports cannot mask current failures, and report parse errors remain failures. Measure readability or byte reduction separately from runtime.