pfangueiro
UserSelf-healing AI agent framework for Claude Code: 13 specialized SDLC/SSDLC agents, 26 skills, 13 slash commands, 9 hooks, launchd watchdog daemon, observability dashboard. Auto-activating, drift-reconciling, snapshot-protected.
Categories
Indexed Skills (27)
deep-analysis
Runs a long, structured multi-step reasoning chain over the sequential-thinking MCP server, with branching, revision, and explicit hypothesis verification. Reserved for consequential, genuinely open problems whose answer cannot be reached by reading the code — an architecture decision with several defensible options, an unexplained performance bottleneck, technology selection, design trade-offs, or a migration and risk strategy. Because it is slow and token-heavy it is not for questions with one correct answer, factual or syntax lookups, summarizing or explaining existing code, a defect with an obvious cause, or applying a fix already chosen. Prefer /investigate for root-cause work on a reproducible defect, and /diverge when the need is a wide option space rather than one reasoned conclusion.
deep-read
Comprehensive codebase reading engine. Systematically reads actual source code line by line through a 6-phase protocol — scoping, structural mapping, execution tracing, deep reading, pattern synthesis, and structured reporting. Source code is the source of truth. Use when needing to truly understand how code works, not just what documentation claims.
diverge
Structured divergence engine for consequential, open-ended decisions. Generates a wide, deliberately non-anchored option space by spawning isolated parallel sub-agents under different cognitive frames, then runs a separate critic pass that scores, declusters, and deepens the best. The divergent complement to /deep-analysis. Warranted only when a decision is both hard to reverse or wide in blast radius AND genuinely open with several structurally different answers — architecture and platform strategy, design-space exploration, de-anchoring a premature conclusion, or a high-stakes what are our options. Triggers include diverge, brainstorm options, explore alternatives, option space, divergent thinking. Because it fans out many agents in parallel it is not for cheap or adjacent asks — naming, wording, picking between two known libraries, or any question with one defensible answer should be answered directly.
execute
Orchestrated task execution engine. Decomposes a large goal into atomic tasks, plans their dependencies, selects an agent, tool, or MCP server per task, runs them in parallel batches, tracks progress, and reports what changed. Warranted only for genuinely multi-step work — roughly three or more atomic tasks spanning multiple files, agents, or tools that benefit from dependency planning and parallelism, such as execute this, orchestrate this, decompose this goal, or build a feature end to end. Because it plans and batches before acting, it is not for a single edit, a one-file change, a question, a lookup, or anything one or two tool calls already resolve — do those directly.
investigate
Deep root cause analysis engine for defects whose cause is genuinely unknown. Runs an 8-phase diagnostic protocol — observe, reproduce, trace, hypothesize, prove, root cause, fix, prevent — using sequential-thinking MCP, multi-pass code reading, git forensics, competing hypotheses, and the 5 Whys, proving the cause with evidence before any fix. Returns a proven root-cause report; the fix runs only after that diagnosis is approved. Use when the user runs /investigate, or when a bug, crash, exception, stack trace, regression, flaky or intermittent failure, data corruption, memory leak, or unexplained slowness needs its root cause found and nobody knows why. Expensive — do not select it when the cause is already known, for a typo, syntax, import or config mistake, for a feature request or refactor, for "what does this code do", or for applying a fix the user already chose; handle those directly. For a live production outage, route to incident-commander first and investigate after service is restored.
skill-creator
Guide for creating effective skills. This skill should be used when users want to create a new skill (or update an existing skill) that extends Claude's capabilities with specialized knowledge, workflows, or tool integrations.
api-contract-testing
Tools and patterns for API contract testing using OpenAPI, JSON Schema, and contract-first development
browser-testing
Hand-driven end-to-end browser testing through the Playwright MCP server — navigating pages, filling forms, asserting HTTP responses, capturing screenshots for visual regression, checking responsive breakpoints and cross-browser behaviour, and recording codegen sessions. Use when a browser genuinely has to be driven interactively and no runnable suite covers the need — authoring e2e, screenshot, or visual-verification coverage where none exists, debugging a failure a suite already surfaced, or exploring a live page. It does not cover running a repository's existing suite. When the repo has a playwright.config file, a tests or e2e directory, or a test:e2e script, running that suite is the correct action and this skill is not required.
ci-cd-templates
Production-ready CI/CD pipeline templates for GitHub Actions and GitLab CI
code-review-checklist
A ten-category human code-review rubric — correctness, design, readability, testing, security, performance, documentation, dependencies, error handling, and style — plus process guidance on feedback tone, review size, and a pre-submit self-review pass. Use when a review verdict is actually being produced on a concrete change, such as reviewing a pull request or diff, self-reviewing before opening one, or defining a team's code review standards. Triggers include code review, review this PR, pull request review, review checklist, review standards. It is a rubric rather than an analyzer, so it does not apply to merely reading or explaining code, answering how something works, debugging, refactoring, or running linters, formatters, and type checkers.
context-escalation
Explicit-level policy for context management in long-horizon Claude Code sessions. Five escalating responses to context pressure — apply the cheapest intervention that still preserves the work. Auto-activates on context limit, near context limit, conversation getting long, /compact, handoff, /save-session.
deployment-runbook
Deployment procedures, health checks, and rollback strategies. Use this skill when deploying applications, performing health checks, managing releases, or handling deployment failures. Provides systematic deployment workflows, verification scripts, and troubleshooting guides. Complements the devops-automation agent.
docker-deployment
Production-ready Docker configurations, multi-stage builds, and deployment best practices
experiment-loop
Autonomous experimentation pattern for iterative code improvement. Describes the modify-commit-run-evaluate-keep/discard loop generalized from autoresearch. Auto-activates on optimize, experiment, improve iteratively, benchmark, metric-driven, A/B test approaches, autonomous improvement, iterate until better.
git-workflow
Git workflow conventions and conflict/recovery procedures — Conventional Commits format, branching models (Git Flow, GitHub Flow, trunk-based), pull request structure and review checklists, merge-versus-rebase choice, cherry-pick, interactive rebase, and guarded use of history-rewriting or destructive commands. Covers the rebase inversion where --ours is the upstream and --theirs is the user's own work. Use when resolving a merge or rebase conflict, choosing or documenting a branching strategy, writing a commit message or PR description to a convention, deciding between merging and rebasing, undoing or reverting a commit, cleaning up history before a PR, or recovering work after a bad reset, rebase, or force-push. Not needed for routine single commands whose syntax is already known — git status, git add, git diff, git log, staging, a plain commit, or a straightforward push or pull.
handoff
Create HANDOFF.md files for session continuity between conversations. Use when ending a session, switching context, or before /compact.
infrastructure-as-code
Infrastructure as Code decision framework, testing pyramid, CI/CD for infrastructure, state management, module composition, policy-as-code, and drift detection. Covers Terraform, CloudFormation, CDK, Pulumi patterns. Auto-activates on Terraform, CloudFormation, CDK, Pulumi, infrastructure as code, IaC, module, state, tfstate, HCL, stack, drift.
kubernetes-ops
Kubernetes operations, manifest generation, Helm charts, Karpenter autoscaling, GitOps (ArgoCD/Flux), network policies, pod security, and troubleshooting. Auto-activates on Kubernetes, K8s, kubectl, pod, deployment, service, ingress, helm, karpenter, eks, gke, aks, node, cluster, namespace, HPA, autoscaling.
library-docs
Quick access to up-to-date library documentation using MCP. Use this skill when you need to reference official documentation for libraries, frameworks, or APIs. Leverages the context7 MCP server to fetch current docs for React, Next.js, Vue, MongoDB, Supabase, and hundreds of other libraries. Complements the documentation-maintainer agent.
multi-agent-orchestration
Advanced multi-agent patterns using Claude Code's built-in orchestration. Teams, background agents, coordinator pattern, worker restrictions, SendMessage protocol, fork subagents. Auto-activates on multi-agent, orchestrate, team, coordinator, parallel agents, worker agents, background agent, agent swarm.
observability-stack
Observability architecture with OpenTelemetry, Prometheus, Grafana, and structured logging. Covers RED/USE metrics, SLI/SLO monitoring, multi-window burn-rate alerting, dashboard design, and distributed tracing. Auto-activates on monitoring, observability, metrics, alerting, dashboard, Prometheus, Grafana, OpenTelemetry, tracing, logging, SLI, SLO.
remote-triggers
Cross-session automation using RemoteTrigger API. Create, manage, and run scheduled remote agents that execute on cron schedules and survive session exit. Auto-activates on remote trigger, scheduled agent, cross-session, persistent schedule, automated agent, trigger API.
scheduled-tasks
Schedule recurring or one-shot tasks using CronCreate/CronDelete/CronList. Auto-activates on schedule, recurring, cron, periodic, poll, remind, every N minutes, timer, interval, check every.
security-scan
Runs a repository or path security scan and produces a severity-ranked report covering hardcoded secrets and credential patterns, dependency CVEs via npm audit and equivalents, OWASP code patterns such as SQL injection, XSS, SSRF, path traversal and insecure deserialization, unsafe file permissions, and missing HTTP security headers. Use when a scan, audit, or vulnerability check is actually being asked for — /security-scan, scan for vulnerabilities, security audit, check for leaked secrets — or before shipping new code that adds authentication, cryptography, input handling, SQL, or file uploads. Reading, explaining, or editing code that merely happens to touch auth or SQL does not warrant it, since it emits a full scan report rather than an inline observation.
sre-runbooks
SRE operational runbooks, on-call handbook, blameless postmortem templates, SLI/SLO/error budget definitions, toil reduction framework, capacity planning, and chaos engineering playbook. Auto-activates on runbook, on-call, postmortem, error budget, SLO, SLA, capacity planning, chaos engineering, incident review, toil, game day.
ui-guidelines
Ant Design house style for one specific React/Next.js application — its design tokens, color palette, typography scale, 8px/12px/16px/24px spacing, animations, and copy-ready patterns for antd tables, forms in modals, selectable card grids, loading states, avatars, and status tags. Use when building or restyling a form, table, modal, card, list, or page layout in a codebase that already standardizes on Ant Design and the new component must match the existing design system. It is not a general UI or accessibility reference, does not apply to Tailwind, Material UI, Chakra, or plain-CSS projects, and carries no chart or data-visualization guidance — charts, graphs, dashboards, and palette choice for data viz are out of scope.
worktree-workflow
Git worktree isolation via the EnterWorktree and ExitWorktree tools — creating a separate checkout on its own branch, working in it, then keeping or discarding it. Use ONLY when a worktree is explicitly requested, by the user or by project instructions (CLAUDE.md / memory). Triggers include worktree, EnterWorktree, ExitWorktree, isolated checkout, side branch. Once explicitly requested it fits work needing a second simultaneous checkout — a risky refactor that may be thrown away, a hotfix on a release branch while feature work continues in the main tree, or two implementations compared side by side — but none of those authorizes the tool on its own, without the explicit request. It does not cover parallelism in general — parallel sub-agents, parallel tool calls, and concurrent tasks are unrelated, and per-agent repo isolation is instead the Agent tool's isolation parameter set to "worktree" — nor an ordinary branch switch, a single-file experiment revertible with git checkout, or a non-git project.
Bio shown is the top-scored skill's repo description as a fallback — real GitHub bios land in a future update.