# Individual catalog review — batch 02 — 2026-10-23

Reviewed **150 complete canonical single-file trees** on protected base `dee7acce0658c35cb276154892700811affcfdb2`: **90 corrected, 60 retained**. The [individual record](individual-review-150-batch-02-2026-10-23.json) binds every reviewed file, original tree, mode and before/after SHA-256, specific findings, sources and limitations. No runtime instructions were executed to discover behavior.

## Resume and scope

This continues the October campaign. The prior October registers contain 250 distinct IDs; September registers add 33 IDs not in that union. Those 283 previously recorded IDs are excluded, as requested, without asserting that historical review certifies their current bytes. Telegram is reserved to [PR2070](https://github.com/sickn33/agentic-awesome-skills/pull/2070), also excluded. No existing review was repeated.

The exact source inventory contains **2,641 root skill trees and 2,743 nested/top-level entrypoints**. Roots containing no tracked SKILL.md and distribution symlink aliases are not new complete-tree reviews. This batch selects the smallest remaining one-file trees, ordered by bytes then ID; it does not represent the larger executable bundles. Across historical records and this batch, 433 root IDs have recorded reviews; 2,207 further root trees remain in the campaign plan, plus the separately reserved Telegram work. This PR is source preparation only: corrections stay unmerged until the full campaign's PRs are ready.

## Repairs

- Replace nonexistent automatic memory/reliability runtimes, lossless compression and pricing guarantees with concrete bounded manual procedures.
- Verify raw webhook bytes and durable processing boundaries; reject empty secrets, concurrent duplicates and ambiguous write retries.
- Preserve Odoo databases/filestores during restore tests, correct company-rule union semantics, version-specific view syntax, dates, fiscal assumptions and invoicing configuration.
- Correct WCAG AA versus AAA targets, missing UI dependencies, GLSL raymarch hit detection, framework cache/compiler defaults and host-specific tool assumptions.
- Prevent same-file shell truncation, secret-bearing errors and unconditional account/runtime claims; make scanner gates fail on configured findings.
- Preserve unknown evidence in estimation, options analysis, legal drafting, meeting actions and provider/API availability.

## Verification and limits

`npm run validate` passed for all 2,743 entrypoints using the existing Python 3.14 interpreter with PyYAML; the default Python 3.15 initially lacked that dependency. `npm run validate:references`, `npm run security:docs` and six offline example regressions passed. Regressions execute review-controlled fixtures, compare them inertly with the published examples, and use synthetic data/transport. They test HMAC byte/secret binding, missing secrets, HTTP-200 quota messages, malformed payloads, key overrides and redacted failures; no provider was called. The full isolated `pr:preflight` passed on implementation head `c129f34f75e14404aff6dd59978eb93cd4dea8a2`. This report-only receipt update is checked again before PR creation; the PR records final-head evidence.

No live infrastructure, database, payment, GPU/browser, legal/medical/financial result or upstream license certification is claimed. Existing source/license declarations are preserved. CI/Tessl results and exact-head merge authority remain separate; no automated review pass is inferred. Generated mirrors remain at their release snapshot.

## Individual dispositions

| Skill | Disposition | Review evidence |
| --- | --- | --- |
| `create-issue-gate` | corrected | Public issue writes require critical risk; optional gate and authorization boundaries retained. |
| `lightning-architecture-review` | retained | Retained revision-specific protocol review, explicit unverified implementation history and funding/broadcast boundary. |
| `recallmax` | corrected | Removed unsupported million-token, lossless compression, automated memory and pricing claims plus mutable installation. Replaced with concrete source-linked manual handoff, uncertainty and budget verification. |
| `cloudformation-best-practices` | retained | Retained minimal VPC template and explicit change-set, replacement-policy and deployment limitations. |
| `accesslint-scan` | retained | Retained locally pinned CLI prerequisites, session-scoped teardown and selector/source evidence boundaries. |
| `claude-code-guide` | retained | Retained project-preserving CLAUDE.md template, task-language reasoning guidance and actual-host verification. |
| `plan-ledger-tasks-yylo` | corrected | Classified ledger/artifact writes as critical; preserved explicit planning scope and external state boundary. |
| `tool-use-guardian` | corrected | Replaced nonexistent automatic wrapper, lossy schema coercion, sensitive logging and unconditional retries with documented-schema validation, replay-safe attempt caps and unknown-write reconciliation. |
| `test-fixing` | corrected | Replaced assumed make target and host-specific Edit instruction with verified project contract; protect dirty work and assertions, distinguish unavailable/flaky checks. Risk now critical for code changes. |
| `legal-advisor` | corrected | Removed remaining compliant-policy and complete-legal-document claims; output now reviewable drafts and evidence/implementation gaps consistent with qualified-review boundaries. |
| `ui-lint` | corrected | Corrected false Tailwind v4 --text namespace conflict with official font-size guidance; scoped design-system rules to actual project conventions and grep matches to candidate findings; preserved source/license metadata. |
| `lightning-channel-factories` | retained | Read complete guide; retained proposal-versus-implementation distinction, unresolved upstream availability disclosure, signing/funding authorization and concrete protocol recovery review criteria. No implementation certification inferred. |
| `progressive-estimation` | corrected | Added actual PERT expected value/dispersion formula and worked example; removed unsupported research-backed multipliers and automatic percentile claims; specified effort/elapsed units, dependencies, local calibration and explicit distribution assumptions. |
| `socialclaw` | corrected | Removed inconsistent 13-platform count and misleading no-account-setup claim; made workspace-account readiness conditional on actual service state and replaced shell-history credential example with private secret loading. |
| `animejs-animation` | retained | Read entire version-labeled v3 example and v4 boundary, reduced-motion, lifecycle and performance limits; retained because concrete animation sequence is useful without claiming bundled runtime or universal compatibility. |
| `macos-screen-recorder` | retained | Read complete single-file tree; retained explicit absent-source/binary disclosure, reviewed-build requirement, screen/audio scope, authorization, input-event separation and real-track verification. No capture executed. |
| `vps-server-management` | corrected | Replaced default root and unrestricted on-box agent recommendation with authorized inventory, verified host key, least privilege, scoped status/log inspection and rollback/health checks. Removed blanket dismissal of EBADENGINE. Marked library paths as unbundled upstream examples. |
| `markstream-angular` | retained | Read complete integration guide, code, alpha/version boundary and safety notes; retained actual installed-package prop/peer verification requirement and malformed-stream/URL/privacy limitations. No Angular runtime assertion made. |
| `vscode-extension-guide-en` | corrected | Clarified advertised reference documents are not in this tree; replaced universal 5MB claim with VSIX file inspection and added concrete webview CSP/resource/message boundaries. Preserved adaptation attribution. |
| `antigravity-design-expert` | retained | Read entire design direction and motion rules; retained scoped use, reduced-motion, keyboard/native scrolling and measured will-change guidance. This is an aesthetic workflow, not accessibility certification. |
| `anti-sleep` | corrected | Checked local caffeinate manual without starting assertions; changed default to idle-system assertion, display assertion opt-in, clarified AC-only -s and lid/shutdown/power limits. Existing PID-specific cleanup retained. |
| `manifest` | retained | Read full telemetry procedure; retained separation of telemetry and routing, absent CLI/plugin disclosure, private credentials, actual transmission/retention authorization, synthetic trace verification and rollback. No host integration changed. |
| `new-rails-project` | retained | Read all stack, database, background job, frontend and verification guidance; retained historical-version warning and actual peer/schema validation, isolated DB/Redis and destination boundaries. No compatibility or build pass asserted. |
| `news-sentiment-engine` | retained | Read complete briefing workflow and optional external setup; retained inert illustrative output, source/date/interpretation distinctions, third-party review boundary, ignored credential file and charge limits. External collector not executed. |
| `healthcare-job-search` | corrected | Checked provider dataset definitions; removed stale employer/job counts, documented actual percentile meanings and dataset timestamps, distinguished national aggregate endpoints from state-filtered tool results. Preserved source and provider disclosure. |
| `executing-plans` | retained | Read full plan-execution process; retained existing todos, user-selected cadence, independent continuation and repository-workflow precedence. No host-specific tools assumed; verification/blocker limits remain explicit. |
| `event-sourcing-architect` | retained | Read full architecture guide; retained correction-event/privacy-deletion distinction, actual concurrency/lag/idempotency requirements and replay suppression of payment/notification effects. No complete audit or automatic saga rollback claim. |
| `interview-coach` | corrected | Added usable direct coaching process independent of unbundled /coach command package, with role-to-evidence mapping, one-question mock, explicit rubric and fact-preserving STAR revision. Marked 23-command scope as upstream advertisement. |
| `options-flow-analyzer` | corrected | Replaced financial certainty and institutional-intent assertions with explicit heuristic buckets; added timestamp/coverage/pagination, consistent volume-versus-OI ratio formula, zero denominator, stale quote/missing delta and threshold sensitivity handling. Replaced market-looking example with synthetic arithmetic. Attributed commercial/performance claims as unverified upstream claims. |
| `pi-web-search` | corrected | Removed assumed global installation/fallback readiness and mandatory query quotas; required actual schema/provider discovery and evidence-driven scope. Added untrusted fetched content boundary while retaining upstream workflow parameter convention and credential-gated optional API example. |
| `azure-communication-callingserver-java` | retained | Retained legacy-only migration index: credential placeholders, alternative snippets, participant-consent boundary and exact-API/version caveat are explicit. Microsoft current package documentation confirms replacement client names. Complete tracked tree read (one regular SKILL.md); existing provenance/authorship/license metadata preserved, not independently relicensed. |
| `terraform-infrastructure` | corrected | Corrected safe risk label and added concrete plan/apply boundary covering backend/workspace identity, secret state, replacements, saved-plan review and lock handling. Phase-based module/state/CI guidance retained. Complete tracked tree read (one regular SKILL.md); existing provenance/authorship/license metadata preserved, not independently relicensed. |
| `writing-plans` | corrected | Corrected missing superpowers dependencies being presented as mandatory bundled capability, worktree prerequisite and nested Markdown fence. Distinguished planned checks from execution evidence and scaled tests to behavior. Complete tracked tree read (one regular SKILL.md); existing provenance/authorship/license metadata preserved, not independently relicensed. |
| `magic-ui-generator` | retained | Retained optional provider workflow: accepted designs can proceed, licensing and generated-code review are explicit, accessibility and hosted-transmission limitations prevent treating generation as production proof. Complete tracked tree read (one regular SKILL.md); existing provenance/authorship/license metadata preserved, not independently relicensed. |
| `tokenwise` | corrected | Removed unsupported global competitor/statistic claims, absolute telemetry guarantee, billing-error bound and installer model-probe guarantee. Added installed-revision, provider-transmission, cost reconciliation and local-log boundaries while preserving attribution/license. Complete tracked tree read (one regular SKILL.md); existing provenance/authorship/license metadata preserved, not independently relicensed. |
| `cdk-patterns` | retained | Retained CDK construct example and deployment safeguards: actual handler prerequisite, retained state, scoped table grant and synth/diff/environment authorization are explicit. Complete tracked tree read (one regular SKILL.md); existing provenance/authorship/license metadata preserved, not independently relicensed. |
| `security-bluebook-builder` | corrected | Corrected a contradictory unconditional template load: use the external support-root template when available or disclose an independent draft. Added distinction between proposed and implemented controls. Complete tracked tree read (one regular SKILL.md); existing provenance/authorship/license metadata preserved, not independently relicensed. |
| `videodb-skills` | corrected | Removed universal upload and historical free-credit claims, clarified missing setup command, added media-rights, hosted-processing/cost, capture scope/stop, credential and stream-link boundaries. Complete tracked tree read (one regular SKILL.md); existing provenance/authorship/license metadata preserved, not independently relicensed. |
| `mirrord` | retained | Retained explicit staging target/context selection, traffic-steal consent and warning that mirror mode can write databases/consume queues. Installation avoids remote shell piping. Complete tracked tree read (one regular SKILL.md); existing provenance/authorship/license metadata preserved, not independently relicensed. |
| `ui-a11y` | corrected | Corrected WCAG 2.2 AA target-size error (24 CSS pixels with exceptions versus 44 AAA/usability preference), and clarified Radix/role does not prove accessible composition. Complete tracked tree read (one regular SKILL.md); existing provenance/authorship/license metadata preserved, not independently relicensed. |
| `wiki-researcher` | corrected | Corrected dead-code proof criterion: missing static call sites cannot prove non-use where dynamic registration/configuration exists; now requires bounded evidence and uncertainty. Complete tracked tree read (one regular SKILL.md); existing provenance/authorship/license metadata preserved, not independently relicensed. |
| `accesslint-diff` | retained | Retained controlled base/head snapshots, safe checkout preservation, schema-verified CLI usage, selector-drift handling and report-only boundary. Complete tracked tree read (one regular SKILL.md); existing provenance/authorship/license metadata preserved, not independently relicensed. |
| `seo-authority-builder` | corrected | Added concrete evidence rules preventing fabricated credentials, testimonials, badges and case studies; labeled numeric score as internal rubric and linked Google explanation that E-E-A-T is not one ranking factor. Complete tracked tree read (one regular SKILL.md); existing provenance/authorship/license metadata preserved, not independently relicensed. |
| `api-analyzer` | corrected | Removed example inference that POST checkout typically requires body; question now depends on supplied endpoint contract, consistent with existing method-specific caveats. Complete tracked tree read (one regular SKILL.md); existing provenance/authorship/license metadata preserved, not independently relicensed. |
| `shader-programming-glsl` | corrected | Corrected missing fragment precision, host/version assumptions, raymarch false-hit detection and unbounded distance growth, plus incorrect perspective w=1 troubleshooting. Preserved illustrative GLSL workflow. Complete tracked tree read (one regular SKILL.md); existing provenance/authorship/license metadata preserved, not independently relicensed. |
| `emblemai-crypto-wallet` | retained | Retained historical integration index with exact-transaction confirmation, provider/custody verification, seed/key prohibition and timeout reconciliation. Endpoints explicitly require current provider verification. Complete tracked tree read (one regular SKILL.md); existing provenance/authorship/license metadata preserved, not independently relicensed. |
| `instructree` | retained | Retained pinned executable-download review, narrow scans, static-analysis limitations and imported-content-as-data boundary. No instruction file changes authorized by scan alone. Complete tracked tree read (one regular SKILL.md); existing provenance/authorship/license metadata preserved, not independently relicensed. |
| `fda-food-safety-auditor` | retained | Retained qualified-reviewer scope, product-specific critical limits, tentative regulatory conclusion and explicit no AI disposition/restart authority. Current official corrective-actions provision consulted. Complete tracked tree read (one regular SKILL.md); existing provenance/authorship/license metadata preserved, not independently relicensed. |
| `agent-memory-mcp` | retained | Retained immutable repository review before install, dependency/lifecycle inspection, scoped persistence/import boundaries and historical-search-results-as-data treatment. Complete tracked tree read (one regular SKILL.md); existing provenance/authorship/license metadata preserved, not independently relicensed. |
| `odoo-orm-expert` | corrected | Corrected prohibition on Python date/datetime domain values using official ORM guidance; distinguished Date/Datetime types and corrected date_order example to typed UTC datetime bounds for the full year. Complete tracked tree read (one regular SKILL.md); existing provenance/authorship/license metadata preserved, not independently relicensed. |
| `powershell-windows` | corrected | Corrected false Unicode/interpolation restrictions, truthiness-based null checks and catch template that continued past errors; clarified native exit-code handling. |
| `godot-4-migration` | corrected | Qualified blanket static-typing performance advice; retained concrete migration fragments and backup/converter validation boundaries. |
| `detect-ai-text` | retained | Retained provider-score framing, non-prose abstention, explicit upload approval and consequential-decision safeguards; API behavior is described as provider-dependent. |
| `seek-and-analyze-video` | corrected | Made missing wrapper/reference dependency explicit, removed unsupported pricing/timing guarantees and added content-disclosure/persistent-memory boundaries. |
| `ux-feedback` | corrected | Fixed undefined refetch in component fragment; made unbundled design reference explicit and avoided unsafe universal retry/short undo windows. |
| `longbridge` | corrected | Removed unsupported safety guarantees about separate CLI side effects, confirmation enforcement and credential handling; retained read-only task boundary. |
| `penetration-testing` | corrected | Replaced shell-redirection placeholder in local inventory example with concrete help-only nmap commands; retained explicit scope and confirmation gates. |
| `kotlin-coroutines-expert` | retained | Retained structured concurrency, cancellation rethrow, supervisor await caveat and Flow delivery limitations; snippets are explicitly incomplete fragments. |
| `payment-integration` | corrected | Corrected PayPal HMAC claim, unsafe acknowledgment-before-persistence advice and racing deduplication; replaced incident insinuations with test scenarios. |
| `security-automation` | corrected | Marked historical unpinned workflow as a non-production sketch with least permissions/tool prerequisites; supplied missing boto3 import, scoped remediation and rollback boundaries; replaced invalid Checkov code and labeled SOAR pseudocode. |
| `odoo-rpc-api` | corrected | Corrected unauthenticated session endpoint presentation, false XML-RPC attachment restriction, admin example and unsafe mutation retry guidance; raised risk for CRUD. |
| `windows-shell-reliability` | corrected | Removed guarantee that no-incremental cleans stale outputs and automatic admin escalation advice. |
| `mise-configurator` | corrected | Explained mise trust/install execution boundary and labeled old version pins as examples instead of production recommendations. |
| `delegating-to-agents` | retained | Retained explicit delegation authorization, scoped ownership, structured prompts, shell-injection warning and remote-host boundaries; background product descriptions need local verification. |
| `optim-agent` | retained | Retained bounded budget/baseline/trial-history procedure, failed-trial accounting and held-out validation; no global-optimum guarantee. |
| `moatmri` | corrected | Resolved inverted moat scoring that would otherwise raise aggregate risk for strong protections; required evidence/unknown treatment and heuristic labeling. |
| `agent-qa-debug-fix` | retained | Retained evidence-to-patch workflow, classifier skepticism, narrow rerun, side-effect approval boundary and no-defect-concealment rules. |
| `ux-copy` | corrected | Clarified missing upstream references, corrected cancel/close semantics and avoided account enumeration/ephemeral-only recovery. |
| `acceptance-orchestrator` | retained | Retained explicit state machine, evidence-backed acceptance, iteration stop conditions and deployment authorization boundary; repository sub-skills must be available. |
| `container-hardening` | corrected | Replaced angle-bracket image placeholders that the shell interprets as redirection with inert replacement markers; existing ABI, workload and scanner limitations retained. |
| `fda-medtech-compliance-auditor` | corrected | Made example severity provisional and replaced universal defect-to-clinical-risk mapping with evidenced safety assessment. Existing QMSR effective-date caution verified against FDA. |
| `postgresql-optimization` | corrected | Clarified that EXPLAIN ANALYZE executes queries and added baseline, locking, maintenance-window and rollback boundaries. |
| `odoo-migration-helper` | corrected | Removed unsupported blanket field/API/chatter migration claims, false pre-manifest assertion, irrelevant npm validation and mandatory intermediate-upgrade rule; supplied target-branch comparison and staging/recovery procedure. |
| `software-architecture` | retained | Retained: scoped architecture/style guidance separates domain concerns, names dependency tradeoffs and provides limitations; no concrete correctness/safety defect requiring a change. |
| `meeting-distiller-pro` | corrected | Prevented invented owners/deadlines, required evidenced attribution and exact quotes, and added transcript privacy/sharing boundaries. |
| `andrej-karpathy` | retained | Retained: complete behavioral guide has concrete scoped examples, assumptions, surgical-change rules and verification limits; upstream attribution/license metadata preserved. |
| `find-bugs` | retained | Retained: exact refs, complete diff inspection, applicability/unverified checklist statuses and evidence-backed findings are concrete; review-only/fix authorization is explicit. |
| `maxia` | corrected | Fixed unterminated execute curl payload quote, removed false connected-state claim, classified marketing claims as unverified, and added explicit transaction/spend/credential/data boundaries. |
| `odoo-qweb-templates` | corrected | Corrected t-esc alias/version and expression guidance, required escaping/trust review and identified custom model/access prerequisites. |
| `odoo-xml-views-builder` | corrected | Corrected v17 kanban-box/chatter examples and view-name advice; clarified unsupported attrs, version targeting and separate access controls. |
| `bdistill-behavioral-xray` | retained | Retained: workflow is concrete and already limits self-probes, usage, hidden-reasoning claims, setup authorization and compliance inference. |
| `odoo-accounting-setup` | corrected | Removed reversed/fixed discount account codes, required tax-reduction configuration and testing, replaced Europe/zero-tax assumptions with localization-qualified eligibility/reporting, and removed unverified lock-date edition claim. |
| `nextjs-best-practices` | corrected | Replaced stale cached-by-default fetch advice, added explicit cache/version model check and authorization/shared-cache boundaries, and qualified Edge choice. |
| `obsidian-cli` | retained | Retained: provides targeting, help/version discovery, read-back and authorization limits for writes, eval, reload and exports. |
| `squirrel` | corrected | Limited strike-three reversion to identifiable own changes and required preservation of unrelated work. |
| `odoo-backup-strategy` | corrected | Replaced production drop/filestore deletion restore with new isolated targets; added failure handling, privacy, consistent dump/store interval and recovery-tested retention; classified final backup/restore risk as critical. |
| `to-prd` | retained | Retained: conversation synthesis/template/testing seams and publication workflow are concrete, with explicit external-action approval limitation and preserved attribution. |
| `fp-taskeither-ref` | retained | Retained: lazy execution and fallible-boundary limitations are explicit; examples validate HTTP/schema and explain ApplyPar/cancellation limits. |
| `verify-citations` | retained | Retained: concrete verification/reporting distinguishes coverage from truth and explicitly requires approval before third-party transmission. |
| `odoo-purchase-workflow` | corrected | Corrected approval equality threshold and manager exception; made partial receipts/bill control and Should Be Paid checks explicit; added operational authorization and qualified service-flow handling. |
| `git-pr-review` | corrected | Removed conditional-diff heading that contradicted mandatory final net-diff verification. |
| `laravel-expert` | retained | Retained Laravel architecture, validation, authorization, queue and database guidance; version scope is explicit and no unsafe executable setup is bundled. |
| `keyword-extractor` | corrected | Prevented inferred topics from being presented as extracted keywords; qualified SEO ranking evidence. |
| `kubernetes-deployment` | retained | Retained phase index with existing context/namespace authorization, immutable manifests, rollout/rollback, secret handling and health-evidence boundaries. |
| `odoo-module-developer` | retained | Retained version-qualified scaffold with explicit missing imports/views/security and sensitive patient/chatter access limitations. |
| `pi-custom-model` | corrected | Updated static-catalog/restart assumptions against current Pi docs; qualified model metadata and fallback claims, protected credentials and model verification. |
| `idea-darwin` | corrected | Provided executable-in-conversation fallback for unbundled slash commands and distinguished scoring from validation. |
| `api-documentation` | retained | Retained scoped workflow index with implementation-backed contracts, sandbox requests, secret handling and separate publishing authority. |
| `using-superpowers` | corrected | Bound imperative skill-loading workflow to instruction hierarchy, available tools and user constraints. |
| `server-management` | corrected | Replaced misleading secret-storage dichotomy and separated liveness from dependency readiness. |
| `seo-page` | corrected | Replaced rigid SEO density/length/word-count claims and missing reference with evidence-aware page analysis and primary guidance. |
| `cis-benchmarks` | retained | Retained docs-only benchmark assessment with profile/version selection, authorization, private-output and no-certification limitations. |
| `meeting-notes` | corrected | Corrected example inventing Maria as action owner despite no assignment in source notes. |
| `cloudflare-workers-expert` | retained | Retained illustrative Worker handlers with current-limits pointer, bounded background-work and deployment/remote-binding limitations. |
| `web-security-testing` | corrected | Added missing authorization and non-destructive testing boundaries; labeled historical OWASP taxonomy and raised operational risk. |
| `vexor-cli` | corrected | Reconciled missing installation reference and command syntax, quoted glob, and documented private-code indexing boundary. |
| `skill-check` | corrected | Prevented invented exact validator scores/counts when rules are absent and removed unsupported false-positive estimate. |
| `odoo-security-rules` | corrected | Fixed company rule union bypass, ACL group example, operation flags, active-company meaning and incorrect field-security advice using Odoo sources. |
| `latex-paper-conversion` | corrected | Corrected unsafe TeX extraction/table conversion and compilation assumptions; added source isolation and fixed example fencing. |
| `plan-writing` | corrected | Removed conflicting verification timing, unconditional location rule, implied bundled scripts and unmatched Markdown fence. |
| `odoo-l10n-compliance` | corrected | Corrected CSD versus FIEL password, PAC submission/test setup, invalid universal EU zero-rate recipe, Fixed/percentage confusion and overly broad localization feature guarantees. |
| `privacy-mask` | corrected | Removed blanket offline/anonymization assurance; explained OCR false negatives, optional model downloads and sensitive dry-run logs while preserving original-image handling. |
| `chrome-extension-developer` | retained | Retained concrete MV3 manifest/messaging example with existing sender validation, least privilege, ephemeral-worker and version-specific messaging limitations. |
| `workflow-yylo` | retained | Retained clear separation of validated workflow storage, execution authority and run artifacts, including revision-safe update and no-invented-command behavior. |
| `documentation-templates` | retained | Retained practical README/API/JSDoc/changelog/ADR templates with explicit placeholder, actual-license and confidentiality boundaries. |
| `openclaw-deployment-hardening` | corrected | Made Trivy examples fail on selected findings, bound scan/SBOM/signing to one supplied immutable image digest, and required a failure-stopping CI shell and trusted signature verification. |
| `unslop-commit` | corrected | Repaired truncated invocation text, disallowed false co-author attribution and made the breaking-change example consistent with route removal. |
| `linux-troubleshooting` | corrected | Added specific trace/log privacy, timing impact, bounded diagnostic and service-restart recovery boundaries to the otherwise useful phased workflow. |
| `blueprint` | retained | Retained cold-start plan procedure, authorization-bound optional delegation, explicit memory boundary and separate immutable-revision installation inspection. |
| `e2e-testing` | retained | Retained scoped Playwright phases and existing test-account, artifact privacy, reviewed baseline, real-browser/device and sandbox limitations. |
| `odoo-sales-crm-expert` | corrected | Corrected predictive scoring disable advice, unsupported Rotting Days assumption, commissions exclusion and geographic-assignment restriction; added version/company/write scope. |
| `doc2math` | retained | Retained grounded extraction schema, explicit missing-information and tagged-inference rules, plus optional hosted BYOK data/account boundary. |
| `screen-adverse-media` | corrected | Added current-service contract prerequisite, removed blanket corroboration and legal-mandate assertions, retained identity ambiguity, original-source corroboration and human decision safeguards. |
| `odoo-shopify-integration` | corrected | Replaced unauthenticated production-like webhook/order snippets with raw-body HMAC helper and explicit durable idempotency, mapping, tax/currency and draft-order procedure; removed official-connector and stable-SKU assumptions. |
| `verify-document` | corrected | Added service-availability prerequisite and removed advice implying AI-text detection could establish authorship; retained upload approval, original-byte preservation and issuer/human verification requirements. |
| `open-dynamic-workflows` | corrected | Clarified that daemon/bridge/scripts are not bundled, required reviewed immutable upstream installation and lifecycle inspection, and distinguished local execution from actual sandbox isolation. |
| `kubernetes-hardening` | retained | Retained restricted namespace, security context, least-privilege RBAC and policy fragments with existing CNI/DNS dependency, namespace-selector and rollout limitations. |
| `seo-sitemap` | corrected | Corrected inclusive URL and uncompressed size limits, lastmod/deprecated-field claims, missing template assumption and unsupported page-count/word-count safe harbors; added bounded fetch and XML parser boundaries. |
| `unslop` | corrected | Fixed destructive same-input/output redirection instruction, kept cleanup in separate files for review, added CLI failure checks and deterministic local cleanup before replacement. |
| `to-issues` | retained | Retained vertical-slice issue template, dependency ordering, user breakdown approval and explicit prohibition on modifying the parent issue. |
| `us-watchlist-alerts` | corrected | Removed unsupported fixed quote-delay advice; retained upstream installation consent, local-state scope, outbound delivery boundary and no-trading restriction. |
| `odoo-project-timesheet` | corrected | Replaced unsupported project billing/menu and automatic budget-alert assumptions with the documented service-product/sales-order flow; separated validation from invoice eligibility and added a two-entry exclusion check. |
| `find-matching-tenders` | retained | Retained concrete resolve/search/match/provenance request sequence; already distinguishes third-party scores from eligibility and requires primary tender confirmation and private-data approval. |
| `docs-architect` | corrected | Added evidence requirements for architectural rationale/history and prohibited invented benchmark, deployment or security guarantees. |
| `vibers-code-review` | corrected | Removed credential disclosure from commit-message instructions; added collaborator/action review and external disclosure/payment boundaries, conditional service claims and explicit feedback-send authorization. |
| `agent-memory` | retained | Retained pinned review-before-install workflow, complete bundle inspection, explicit storage/import scope and historical-results-as-data handling. |
| `full-output-enforcement` | retained | Retained bounded completeness workflow and working small temperature example; limitations preserve scope, missing evidence, token constraints and legitimate syntax. |
| `bevy-ecs-expert` | retained | Retained query/resource/system examples and explicit version/API limitations, frame-based score caveat and B0001 internal-query distinction. |
| `git-commit-message` | retained | Retained staged-diff grounding, safe multiline command guidance and no-implicit-commit boundary; examples are explicitly hypothetical. |
| `bdistill-knowledge-extraction` | retained | Retained distinction between adversarial self-correction and independent validation, uncalibrated score warning, local/host processing boundary and synthetic medical-data cautions. |
| `linux-hardening` | corrected | Separated pwquality and audit rule-file contents from shell command blocks; preserved access recovery, scoped hardening and audit-secret/retention cautions. |
| `alpha-vantage` | corrected | Corrected WTI commodity identifier, removed quota-evading half-second-delay advice, bounded HTTP requests, checked semantic API errors before indexing and limited outputsize/quote assumptions. |
| `design-orchestration` | retained | Retained bounded design routing with routine-fix exemption, explicit delegation authorization and documented single-agent alternative. |
| `ii-commons` | corrected | Clarified that freshness cutoff does not pin an immutable result set; require retrieval time/query/identifiers and preserved records for reproducibility. |
| `react-patterns` | corrected | Corrected blanket Hook rule for the use API, made Compiler benefits conditional on setup, and replaced blanket effects-to-server-components advice with derived-render/external-sync guidance. |
| `smartui-skill` | corrected | Corrected Playwright snapshot import/package, width-only web viewports and explicit SmartUI config execution; identified unbundled playbook and added cloud page-data and baseline-review boundaries. |
| `baseline-ui` | retained | Retained scoped polish constraints with existing product/accessibility/framework precedence; keyboard/focus primitives, reduced motion, paste and inline errors are concrete. |
| `github-issue-creator` | corrected | Removed invented workflow completion, final success and agent configuration from examples; preserved uncertainty and explicit publication authorization. |
| `modellix` | retained | Retained concrete asynchronous CLI workflow and explicit model/schema/availability limitations, paid submission review, unknown-outcome check and credential/output boundaries. |
| `codex-subagent` | corrected | Replaced universal never-API-key authentication claim with explicit intended-account/mode verification and prohibited silent billable fallback; added delegation authorization boundary. |