AI Agent Workflow Collection

Default to this Skill for an unqualified PDF review and for checks involving searchable text, structure, tables, page-located claims, numeric consistency, citations, scanned pages, OCR, rendering, completeness, or visual layout. Detect available capabilities and degrade within this Skill when tools are unavailable; do not modify the source PDF or claim unsupported verification. Do not use when the user explicitly requests pdf-review-basic or a lightweight, text-only check without OCR, rendering, or complex layout review.

Profile, clean, standardize, transform, fill, or deduplicate spreadsheet or delimited data into an authorized new output with traceable rules, before/after statistics, an audit log, and validation of formulas, encoding, dimensions, and key summaries. Do not overwrite source data, guess critical values, or delete plausible records, and degrade to a cleaning plan when execution tools are unavailable. Use excel-data-quality-check-basic instead when the user requests only a read-only data-quality audit or issue report without changes or a cleaned copy.

Reproduce and locate a specific runtime bug by tracing the first trustworthy failure through code, logs, configuration, existing tests, and minimal experiments into an evidence-backed causal chain. Use for non-CI symptom-to-root-cause diagnosis; do not absorb CI pipeline failures, general code review, broad health checks, requests to add or modify tests, or requests to implement a fix.

Diagnose a CI-specific failure from workflow definitions and logs by locating the failed workflow, job, step, and first trustworthy error, then comparing CI and local environments. Use for CI pipeline context; do not use as a general local bug analysis or implement fixes, reveal credential values, or recommend weakening security controls.

Audit configuration sources, precedence, environment overrides, defaults, required fields, types, validation, drift, obsolete settings, dangerous defaults, undocumented variables, and credential-commit risk using repository evidence. Use for configuration-specific risk; do not perform a broad health check, read or display sensitive values, connect to external environments, or modify configuration.

Compare one or more artifacts for conflicting terminology, figures, dates, units, people, versions, links, and conclusions, reporting both locations, severity, conflict type, and a recommended authority. Use when reconciliation across content is primary; use document-review for holistic quality of one document, and do not select a correct value without authoritative evidence or claim unsupported file-format checks.

Review a concrete database schema or data migration for reversibility, data safety, locking, backfill, replication, compatibility, zero-downtime rollout, validation, and rollback. Use for migration-specific risk; do not perform generic impact analysis, execute migrations, connect to production, or modify migration files.

Verify whether a specific symbol, file, route, configuration path, or feature is safely removable by checking static and dynamic usage, registries, build entries, templates, plugins, scripts, and external contracts. Use for evidence of unused code; do not perform general impact analysis, assume absence from text search proves safety, or delete code.

Analyze a specific dependency upgrade using manifests, lockfiles, repository usage, and verified official release evidence, including breaking changes, runtime requirements, security, licensing, ecosystem compatibility, validation, and rollback. Use before changing dependency versions; do not perform general impact analysis or modify manifests or lockfiles.

Convert supplied meeting notes or transcripts into a summary, decisions, action items, owners, due dates, dependencies, and open questions while distinguishing discussion, proposals, decisions, and commitments. Use when the source is meeting material; do not invent owners, dates, decisions, or task-system updates, and mark missing fields TBD.

Analyze a specific performance regression against a defined baseline and workload using profiles, benchmarks, traces, logs, or reproducible experiments, covering algorithms, I/O, N+1, caching, memory, concurrency, network, database, and rendering. Use for measured regression depth; do not replace general code review, infer slowness from appearance alone, or optimize code.

Design an audience-specific presentation story and slide-by-slide outline with each slide's purpose, title, key points, evidence, and suggested visual. Use when the default deliverable is an outline; generate PPTX only when the user explicitly requests it and capable presentation tools are available, and do not invent missing data or turn a report into dense slides.

Perform an evidence-backed, read-only release gate for a specific candidate version and return GO, GO WITH CONDITIONS, or NO-GO after checking workspace state, versioning, changes, validation, compatibility, rollback, documentation, and sensitive artifacts. Use for release candidates; do not use as a broad project health check or execute release, commit, tag, push, or version changes.

Classify the current repository task and recommend one existing Repo Doctor Skill or an ordered workflow without executing it. Use when the user is entering an unfamiliar repository, is unsure which Skill to choose, or needs the next safe step across clarification, specification, planning, implementation, testing, review, diagnosis, release readiness, or session transfer; do not use for ordinary questions with no repository workflow decision.

Resolve the material product, behavior, compatibility, data, permission, security, migration, and acceptance decisions that remain open before specification or implementation. Use when a request contains consequential ambiguity; inspect repository evidence before asking one focused question at a time, then produce a clarification summary for requirements-to-spec. Do not use for an already clear small change, write the final specification, edit long-term documentation, or implement code.

Turn a confirmed specification, root-cause finding, or impact analysis into an atomic, verifiable, reversible implementation plan with explicit scope and risk controls. Use only when inputs and blast radius are sufficiently known; route ambiguity to requirements-to-spec, unknown impact to change-impact-analysis, architecture rationale or decision documentation to architecture-decision-record, and execution to safe-fix-implementation.

Review a code change or PR broadly for correctness, compatibility, security, maintainability, and tests. Use security-focused-review, performance-regression-analysis, or api-contract-review when the user requests one bounded specialist assessment; do not implement fixes.

Implement one small, safe, verified production fix after a clear diagnosis. Use safe-test-implementation for test-only changes and documentation-sync for documentation-only changes.

Add the smallest set of high-value tests from a test-gap analysis, confirmed behavior, or verified fix, following the repository's real framework and conventions. Use to edit tests, fixtures, and necessary test helpers only; do not use for general bug fixes, production refactors, or coverage-only assertions, and stop before production-code changes unless the user explicitly expands scope.

Perform a scoped security review by establishing assets, trust boundaries, attacker prerequisites, and evidence-backed findings across authentication, authorization, validation, injection, path handling, SSRF, XSS, CSRF, deserialization, credential handling, logging, and dependencies. Use for security depth; do not replace general code review, run attacks, access production, reveal credentials, or implement fixes.

Perform a strictly read-only, pre-release quality audit of one AI Skill, a Pack, a plugin, or a Skills repository across structure, triggering, workflow, progressive resources, bilingual cross-platform output, safety, and publishing integration. Use when maintainers want findings, severity, evidence, and a release recommendation; do not audit ordinary application code or PRs, run a broad project health check, gate a product release candidate, or automatically fix files.

Decompose a confirmed specification, implementation plan, or settled conversation into independently executable and verifiable vertical work items with dependencies, safe parallelism, conflict areas, tests, risks, and rollback notes. Use for delivery planning of a multi-behavior scope; default to a local Markdown plan and do not create external issues. Do not use for unresolved requirements, a single small fix, or a file-by-file technical-layer checklist.