Agent Skills

A collection of reusable skills for AI coding agents. Each skill is a self-contained package that enhances an agent's capabilities in a specific domain.

Behavior-preserving code simplification and readability-focused refactoring for day-to-day coding work. Helps turn dense, nested, or inconsistent code into clearer local code without broadening scope or inventing unnecessary abstractions.

Key capabilities:

• Supports both review-only and direct-edit workflows for cleanup tasks
• Prioritizes high-value readability wins such as flattening nesting, removing dead code, clarifying naming, and reducing needless indirection
• Loads language-specific guidance for TypeScript/JavaScript, PHP, Python, Go, Rust, and shell scripting from references/
• Treats behavior preservation, scope discipline, and validation status as explicit parts of the skill contract

Pressure-testing skill for product, architecture, scope, and sequencing decisions before reassurance. Helps an agent challenge plausible ideas first and endorse them only after the choice survives scrutiny.

Key capabilities:

• Challenges before agreement instead of leading with reassurance
• Uses named decision lenses to test evidence, scope, timing, complexity, reversibility, and opportunity cost
• Surfaces the strongest counterarguments first and states what evidence, trigger, or constraint would change the call
• Ends with a forcing question and gives explicit follow-up guidance after the user responds

Apply Patrick Winston's MIT presentation framework to craft compelling talks, audit slides, make ideas memorable, structure persuasive presentations, and design teaching props and stories.

Key capabilities:

• Implements Winston's complete framework across 10 frameworks and 3 operating modes (Build, Audit, Coach)
• Covers empowerment promises, Star framework, four heuristics, board vs. slides, slide crime audits, props & stories, and how to stop
• Provides structured talk planning, slide auditing with 10 slide crimes, and delivery coaching
• Includes misinterpretation guards for commonly misquoted Winston advice

Platform-aware OpenShift cluster health diagnostics for control plane, operators, nodes, MCPs, and key platform subsystems across bare metal, virtualized, cloud, and SNO environments. Emphasizes read-only investigation and evidence-based severity classification.

Key capabilities:

• Performs structured health checks across cluster version, operators, nodes, MCPs, etcd, authentication, ingress, DNS, networking, storage, monitoring, registry, console, and certificates
• Detects platform topology and infrastructure type (for example BareMetal, VSphere, AWS, Azure, GCP, SNO) and adapts checks accordingly
• Classifies findings into Healthy, Warning, and Critical with explicit blast-radius and impact guidance
• Distinguishes quota/app issues from platform-level failures for pending/crashing pods to avoid false escalation
• Produces actionable output with executive summary, evidence, priority actions, and uncertainty notes

Fetch and display GitHub PR review comments for the current branch in the code review UI, enabling inspection of feedback before deciding how to respond.

Key capabilities:

• Verifies gh CLI authentication before making API calls
• Fetches issue comments, diff comments, and reviews via the GitHub API with pagination support
• Trims large diff hunks to a focused window around commented lines
• Supports both script-based fetching and manual fallback commands
• Renders comments via insert_code_review_comments with proper location and reply metadata

Version-discovery and upgrade-path skill for OpenShift releases using Red Hat APIs. Supports unauthenticated public upgrade graph queries and authenticated metadata queries for detailed lifecycle and managed-service availability information.

Key capabilities:

• Discovers currently active OpenShift minor and latest patch versions dynamically (without hardcoded release assumptions)
• Queries channel-specific release availability (stable, fast, candidate, eus) across architectures
• Computes valid one-hop upgrade targets from a given current version using the public upgrade graph
• Supports authenticated clusters_mgmt lookups for ROSA/HCP enablement flags and end-of-life metadata
• Provides script-based and raw API workflows for both human-readable and JSON-parseable outputs

Senior-level production resilience and failure-mode review for code, services, and system designs. Acts as a hybrid Staff SRE, Principal Engineer, and Incident Commander — finding material production failure modes and providing evidence-calibrated fixes with priority rankings.

Key capabilities:

• Reviews code through twelve failure lenses: dependency failure, load & concurrency, network & latency, data freshness & consistency, retry & backpressure, debuggability, observability & alerting, change management & rollback safety, fault domains & disaster recovery, security & abuse as reliability, quota & limit exhaustion, and complexity tax & architecture fit
• Calibrates severity using impact, likelihood, blast radius, detectability, recoverability, and existing controls
• Provides two review modes: Quick (top risks, fast pass) and Full (deep analysis with validation and monitoring plans)
• Applies heightened resilience scrutiny when code is identified as AI-generated without inferring authorship from code smells
• Ships with reference checklists for all twelve lenses, severity calibration, and validation/monitoring patterns

Risk-first QA skill for requirement tracing, test planning, defect reproduction, regression control, and evidence-based release decisions.

Key capabilities:

• Six operating modes: review, test-plan, execute, regression, bug-hunt, and mode-selection with explicit rules
• Builds risk-based test plans organized by category (happy path, boundary, negative, error handling, concurrency, security, state transitions, data integrity, compatibility, performance)
• Exploratory testing guidance for unexpected inputs, realistic data volumes, adversarial payloads, and UI states
• Structured bug reports with severity calibration, reproduction steps, and evidence
• Test quality standards covering determinism, speed, readability, isolation, maintainability, and trustworthiness
• Special considerations for AI-generated code with 11 blind-spot signals

Each skill follows a consistent package layout:

skill-name/
├── SKILL.md          # Skill definition (frontmatter + instructions)
├── package.json      # Name, version, description, keywords
├── VERSION           # Current version
├── CHANGELOG.md      # Version history
├── references/       # Deep-dive reference materials
└── tools/            # Validation and utility scripts

Install the development/CI dependencies, then run the repository validator and all root/per-skill tests:

python3 -m pip install -r requirements-dev.txt
python3 scripts/validate_skill_collection.py
python3 scripts/run_test_suite.py

The collection validator uses the official skills-ref library for Agent Skills frontmatter and naming validation, then applies repository-specific Markdown and script checks. The test runner executes each top-level suite in a separate pytest process because independently packaged skills can intentionally reuse helper module names.

For a package-specific validation pass:

cd production-resilience-reviewer
bash tools/validate_skill_package.sh

Generated .skill bundles are treated as build artifacts, not source files. Build them locally into dist/ with:

python3 scripts/build_skill_artifacts.py

CI also builds these bundles and publishes them as workflow artifacts. Tagged v* releases attach the generated .skill files to the GitHub release.

See individual skill directories for license information.