assurance

Plugin assurance from marketplace greenblacked-ai in greenblacked/AI. Commands read from the repo's manifest — 2 days ago.

Skills (11)

• audit-trail-design

  Design the audit log for user and admin actions: start from the investigation questions (who did what to which record, when, from where, allowed or denied), log denials as well as successes, write state changes in the same transaction as the change, keep the trail in an append-only store separate fr

• control-evidence

  Collect SOC 2 or ISO 27001 audit evidence without a screenshot scramble: map each control to one system-generated artefact, with an API export preferred over a screenshot; define every population and the sampling period before the auditor asks, reconciled against an independent source; automate date

• dpia

  Run a data protection impact assessment as a dated, versioned procedure rather than a form filled in after launch: screen against a written trigger list first and record the answer either way, describe every data flow, test necessity and proportionality field by field, list risks to the people whose

• fuzz-harness

  Put a coverage-guided fuzzer on your own parsing and decoding code: one small harness per entry point that takes bytes and must not crash, hang or exhaust memory, a seed corpus from real files and past bug reports, builds under AddressSanitizer and UndefinedBehaviorSanitizer, every crash minimised,

• licence-compliance

  Audit the open-source licence obligations of what you ship: inventory every component from lockfiles and built artefacts, with transitive packages, vendored code, fonts and base images included; classify each artefact by how it reaches people — hosted, sent to browsers, shipped binary or image, publ

• load-testing

  Run a load test that answers one stated question and cannot pass by accident: fix the question and SLO-based pass thresholds before the run, take the workload mix and parameters from production logs, drive an open arrival-rate model, not fixed virtual users, so coordinated omission cannot hide stall

• mutation-testing

  Measure whether the tests fail when the code is wrong, not merely whether they run it: run a mutation tool on only the lines a pull request changed, triage every surviving mutant as an equivalent mutant, a missing assertion or an untested branch, kill the worst survivors by writing the test that not

• property-based-testing

  Replace hand-picked example lists with properties a library checks across generated inputs: choose the property class first (round trip, idempotence, oracle comparison, an invariant after any sequence of operations, never crashes), build generators that construct valid domain objects instead of filt

• risk-acceptance-register

  Record knowingly accepted risk so the acceptance expires instead of becoming permanent: each entry states the risk, the compensating control and how it is verified, an accountable owner above the engineer who proposed it, and an expiry date; every tool-level suppression points to an entry; owners ar

• test-order-bisect

  Find the polluter behind a test that passes alone and fails in the full suite, then remove the leak rather than the symptom: replay the failing order from its recorded random seed, delta-debug the tests that ran before the victim by halving until one polluter and one victim remain, name the shared s

• traffic-replay-diff

  Prove a rewrite, port or migrated service answers the same as the old one by replaying captured production requests offline against both and diffing responses: capture sanitised read traffic, start both from one data snapshot with identical stubs, replay old against old first to measure noise, norma

Subagents (1)

• replay-diff-reader

  Read a traffic replay's raw response-diff output (the old-against-old noise runs, the candidate-against-old diffs and the request index) and return it bucketed by signature, each bucket classed as noise, candidate bug, pre-approved or needs sign-off with one example request, so the diff files never

Similar plugins