security

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

Skills (31)

• access-review

  Review who and what can do what, and reduce it toward least privilege without breaking production: choose between a scheduled review of the whole estate and a targeted reduction of one identity, inventory humans, service accounts, workload identities, CI/CD and third-party grants, judge on last-acce

• agent-extension-audit

  Vet a third-party agent extension before installing it: a skill with scripts, a Claude Code or other plugin, a hook, an MCP server, a slash command. Fetch it to a quarantine directory without running anything, separate what executes automatically from prose, read every executable file in full, list

• agent-security-review

  Review an LLM agent where untrusted web pages, retrieved documents, files, messages or tool results can influence privileged tool calls: trace content through the model into action parameters; require tool-side authorization and tenant scoping at execution; bind human approval to the immutable norma

• analysis-efficiency

  Cut the cost of an analysis-bridge or decompiler-backed investigation without losing evidence: pin the analyst question and a correctness oracle first, measure one representative task, find the earliest repeated engine work, prefer summary results and retained Evidence paging over full reports and r

• apk-analysis

  Analyse an authorised APK, AAB or split set from the shipped package: record the target digest, read signing identity and certificate lineage before the code, inventory the manifest's permissions, exported components and receivers, take a summary-first dex overview, then follow strings and method re

• app-bundle-analysis

  Read a signed app package you are authorised to examine — a desktop .app bundle, a mobile IPA, or an app inside a DMG — when the source is missing or does not match what ships: identify the bundle from its Info.plist, read the signature and entitlements, follow the frameworks to the runtime family (

• auth-design

  Design or review how a system proves who a caller is and keeps that proof safe: whether to build identity at all, authorization code with PKCE per client type (the implicit and password grants are removed, not discouraged), exact-string redirect URI matching, state, nonce and PKCE, session id rotati

• authorization-design

  Design or review application authorization: who may perform which action on which resource, in which tenant and under what conditions. Use when defining RBAC, ABAC or ReBAC; building a permission matrix or policy engine; preventing IDOR/BOLA and cross-tenant access; placing server-side checks on rou

• authorized-device-debloat

  De-bloat or cut ads on a smart TV, streaming box or similar APK-based device the operator owns or is authorised to modify: identify ad and bloat packages from Evidence — package names, receivers, observed endpoints, APKs pulled from that device — plan removal as reversible steps (pm disable-user bef

• certificate-automation

  Move a certificate estate onto automated ACME issuance and renewal ahead of the CA/Browser Forum's shrinking maximum lifetimes. Inventory from outside first — CT logs and endpoint scans, not only what the team tracks — because the one nobody owns is the risk. Classify endpoints as native-ACME, cert-

• cil-assembly-analysis

  Analyse a managed CIL (ECMA-335) assembly you are authorised to examine — a PE .dll or .exe, a Mono or Unity Managed assembly, or a library shipped in a managed package — when the source is missing: identify the CLI header and target framework before decompiling, list namespaces and types, decompile

• compromise-evidence-preservation

  Preserve evidence from a suspected compromised host, instance, container or account before containment destroys it: record contain-now versus observe-first, follow order of volatility (memory and live connections, processes, disk, then logs), isolate rather than destroy (quarantine group, standby, c

• credential-stuffing-response

  Detect, size and shut down a credential-stuffing campaign against login without locking customers out: confirm from the data that it is stuffing (many accounts per source, few attempts each, low non-zero success) and not spraying or brute force, size it by accounts per IP, success rate by fingerprin

• data-privacy

  Handle personal data so the obligations are met before anyone tests them — settle whether you are controller or processor and under which regime first, because that decides whom you tell and on what trigger; classify what is held and why, because retention, access and deletion all follow the class;

• dependency-triage

  Work a queue of dependency vulnerability alerts — Dependabot, npm audit, pip-audit, govulncheck — into a decision each: reachability before severity, CVSS base versus environmental, EPSS and CISA KEV as the two signals that mean today rather than this quarter, direct versus transitive and the three

• detection-engineering

  Write, test and retire security detection rules. A rule is accepted only with a one-sentence hypothesis naming attacker behaviour, a log source and field verified populated, a Sigma or native rule in version control, a positive test that replays an emulated event, a 30-day backtest with measured ale

• firmware-analysis

  Extract and map a firmware image you are authorised to examine — a vendor update file, a flash dump or a device image — when there is no source and the question is what the device actually runs: digest it before touching it, identify the container from magic and entropy rather than the filename, inv

• host-hardening

  Audit or harden a Linux or Unix host — a VM, a bare-metal server or a bastion — against a named baseline: preserve a second way in before touching sshd or sudo, export the config either side, change one class at a time on a canary host, and verify over a path you did not edit. Covers SSH and keys, s

• iac-review

  Review an infrastructure-as-code change against the machine-readable plan JSON rather than the human-readable plan text — blast radius and destroys first, then replacement causes, drift, state moves, secrets in state, pinning, the policy-as-code gate, cost, and a verdict. Also covers Ansible review

• image-hardening

  Build or audit hardened container images — base image selection down a scratch, distroless, Chainguard, alpine, slim ladder; digest pinning; multi-stage builds; numeric non-root UIDs; BuildKit secret and cache mounts; exec-form entrypoints; multi-arch buildx; SBOM generation; a vulnerability severit

• intrusion-timeline

  Reconstruct what an attacker or unknown actor did, in order, from audit logs, and bound the blast radius: anchor on the earliest confirmed attacker action then work backwards to initial access and forwards to the last action, normalise sources to UTC with measured clock skew, chain each AssumeRole h

• native-function-matching

  Compare two builds of the same native target you are authorised to examine — a patch release, a vendor update, or a version you have and one you do not — and say which functions are the same, which moved, which changed, and which are new or gone. Establishes identity from digests and build ids, inve

• packaged-desktop-app-analysis

  Open up a packaged desktop application you are authorised to examine when its source is missing: identify the packaging first — Electron with an ASAR bundle, a Node single executable, PyInstaller, or a bundled JVM or managed CIL app — read the application's own JavaScript or bytecode out of the cont

• reverse-engineering

  Reverse engineer a native ELF, PE or Mach-O binary, Unity IL2CPP build, mobile app, JVM class file, smart TV .ipk or web bundle you are authorised to examine, for malware triage, CTFs, vulnerability research or interoperability: identify the format, read static structure (strings, symbols, bytecode,

• secret-rotation

  Rotate a credential, or contain one that has leaked, without an outage or a leftover copy: choose the planned or incident path first, inventory every consumer from access logs not the runbook, issue a second live credential not an in-place edit, prove adoption from last-used telemetry past reporting

• security-review

  Review a change for the vulnerability classes actually exploited, in order: establish what the diff reaches and who can reach it, object-level authorisation rather than route-level, injection across SQL, command, template and LDAP, deserialisation, SSRF, secrets reaching a log or repo, unsafe defaul

• security-test-engagement

  Commission or run an authorised penetration test, red-team exercise or bug-bounty engagement from the asset owner's side: a signed authorisation naming the exact scope, systems, IP ranges and dates before any testing starts; rules of engagement with explicit stop conditions and a named emergency con

• tenant-isolation-testing

  Build a harness that proves one tenant cannot read or affect another's data at every layer, not only the API: two sentinel tenants with greppable markers planted in rows, files, search, cache, queues, object storage, exports and logs; a cross-tenant matrix generated from the API description that ass

• threat-model

  Model what could go wrong with a design before the code exists — draw the data flows and mark every trust boundary first, because a threat lives where data crosses one; apply STRIDE per crossing rather than per component, because a crossing has two named parties and a payload and a component has nei

• vulnerability-disclosure

  Run intake, triage, fix and publication of vulnerability reports from outside researchers about your own product: publish the front door first (security.txt per RFC 9116 with Contact and Expires, a scoped policy with counsel-reviewed safe harbour, a private channel), then acknowledge, reproduce, sep

• wasm-analysis

  Inspect a WebAssembly module you are authorised to examine when there is no source: confirm it is a WASM v1 module from its header rather than from the filename, list imports, exports, memories, tables and custom sections with WABT, read the embedded names and strings, and decide what the module can

Slash commands (2)

• /security:blast-radius

  Report what a Terraform or OpenTofu apply will actually destroy and replace, destroys first, from the plan JSON rather than the plan text.

• /security:image-audit

  Audit a container image for base-image currency, non-root execution, secrets in layers, and a defensible vulnerability gate, from the image itself rather than the Dockerfile.

Subagents (5)

• extension-reader

  Read a third-party agent extension (a skill, plugin, hook, MCP server package or slash command) from a local quarantine directory, without running anything, and return what executes automatically and when, every executable file with what it does, declared versus effective permissions, network and cr

• pii-reader

  Read a schema dump, a sample of rows, a log excerpt or an exported field inventory and return which fields carry personal data, the category each falls in, and where it flows onward, keeping the sample itself out of the caller's context. Use when the material is bulky enough that reading it inline w

• plan-reviewer

  Read a Terraform or OpenTofu plan as JSON and report what the apply will actually do, destroys first, with the reason each resource is being replaced. Use when a plan, tfplan or plan.json needs risk-assessing before approval, or when someone asks why a resource is being recreated.

• policy-auditor

  Read IAM policies, role definitions and access logs and return least-privilege findings — the gap between what an identity is permitted to do and what it actually did, plus wildcards, privilege-escalation paths, unconditioned third-party trust and standing admin. Use when an access review is due, wh

• supply-chain-exposure-reader

  Given a compromised package, action SHA or version window, walk lockfile git history and CI run logs across every repository in scope and find which repositories, workflow runs and jobs actually resolved or executed the bad version — not merely whether today's lockfile still contains it — and which

Similar plugins