convez.ai
SkillAutomaçãoIntermediário
conforma

Conforma Agent Skills

conforma/claude-skills

Skills modulares para automação de políticas, sprints e vulnerabilidades

Ver no GitHub
InicialPluginScore · de 0 a 100

Conjunto de skills reutilizáveis baseadas em Markdown para agentes executarem tarefas de política Conforma, workflows de sprint Jira, patching de vulnerabilidades e manutenção de repositórios. Independente de modelo específico, funciona com qualquer harness que suporte agent skills.

Instalação

$npx convez-ai conforma/claude-skills

1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.

Métricas

Estrelas
0
Forks
7
Acompanhando
0
Saúde
62%
Último commit
há 1 dia
Versão
—
Issues abertas
8
Contribuidores
6

Casos de uso

  • Investigar e corrigir violações de política Conforma
  • Automatizar workflows de sprint em Jira
  • Executar patching de vulnerabilidades em repositórios
  • Capturar e reutilizar conhecimento técnico entre subsistemas

Veredito

Coleção bem-estruturada de skills para integração com agentes empresariais, focada em DevOps e governance. Requer configuração específica do harness e compreensão de seus mecanismos de descoberta, mas oferece composição modular sólida.

README

A tradução para o português deste README ainda está na fila.

Conforma Agent Skills

Reusable, Markdown-based agent instructions for Conforma policy work, Jira sprint workflows, vulnerability patching, and repository maintenance. The skill content does not depend on a particular model or agent harness. Harness-specific discovery, invocation, permissions, and MCP configuration remain the responsibility of the client.

Use the skills

Clone the collection:

git clone https://github.com/conforma/agent-skills.git

Preserve the complete skills/ tree, including each skill's references, templates, examples, scripts, and helper files. Install it in a directory supported by your agent harness. There is no single discovery path shared by all harnesses.

If your harness does not discover Agent Skills, give any tool-capable agent the path to the relevant SKILL.md and ask it to read and follow the instructions. For example:

Read /path/to/agent-skills/skills/ec-policy-debugging/SKILL.md and use it to investigate validation.log.

The collection follows the SKILL.md name/description convention. Some clients discover skills automatically; others require a client-specific install location or an explicit file path. Check the active harness documentation for discovery and invocation behavior. Resource links are relative to each skill file, not the current working directory.

When installing a subset, include linked resources and skills: ec-debug-violations uses ec-policy-debugging and ec-setup; ec-policy-debugging links to ec-setup; groom uses the sprint-triage reference; and renovate-act uses the Renovate triage rules reference. Copy the entire vuln-patcher directory, including its scripts/ folder.

Client-specific examples

These examples describe client installation, not alternate skill implementations:

  • Codex (skills documentation): its documented project skill directory is .agents/skills/. Create it if needed, then copy the linked skill directories there, for example mkdir -p /path/to/project/.agents/skills && cp -R /path/to/agent-skills/skills/ec-policy-debugging /path/to/agent-skills/skills/ec-setup /path/to/project/.agents/skills/.
  • Pi: load a specific skill with pi --skill /path/to/agent-skills/skills/ec-policy-debugging/SKILL.md.
  • Claude Code: copy complete skill directories to the project or user skill location supported by your installed version. The former repository layout used .claude/skills/; that client directory is no longer the source layout here.

Client conventions change. Confirm native discovery paths and command syntax in current client documentation. A skill can also be invoked by file path where the harness permits local file access.

Skills

SkillPurposeAdditional requirements
capture-contextRecord reusable subsystem knowledge after implementation.Repository file access
ec-debug-violationsAnalyze Conforma validation logs and investigate violations.Python 3; kubectl for TaskRun logs; policy access tools as needed
ec-policy-debuggingUnderstand a policy rule by inspecting its source, metadata, tests, and input data.Python 3 for the bundled log summarizer; pytest to run its tests
ec-setupPrepare a local reproduction from a Conforma validation log.Conforma CLI and conftest; registry access for OCI policy sources
generate-policyCreate and verify Conforma policy rules, configuration, and tests.Conforma CLI and OPA; image/key/registry access when validating an image
groomAssess one Jira story and, after confirmation, post a review and readiness label.Jira MCP capabilities for reading, commenting, and updating labels
mid-sprint-addRecord a mid-sprint addition and optionally move the issue into the sprint.Jira MCP capabilities
renovate-actRevalidate and carry out user-approved Renovate triage actions.Authenticated GitHub CLI (gh)
renovate-triageCategorize automated dependency-update pull requests and produce a report.Authenticated GitHub CLI (gh); skopeo is optional
scaffold-repoAssess a repository with AgentReady and guide approved improvements.AgentReady CLI
sprint-reportReport mid-sprint additions, with optional Slack posting.Jira MCP; Slack MCP for posting
sprint-triageAssess Jira stories in the current sprint for implementation readiness.Existing MCP tools mcp__work-assistant__jira_sprint_issues and mcp__work-assistant__jira_get_issue
vuln-patcherClassify Conforma vulnerability tickets and prepare verified UBI RPM or Go-module patches.Jira read access; verified Conforma CLI checkout; tools and registry access depend on remediation class. Stops before push or PR creation pending approval.
write-ruleWrite a Rego policy rule and tests from a requirements file.OPA and Conforma policy references

The skills specify their inputs, outputs, and required operations. They do not grant tool access. Configure MCP servers, credentials, command execution, file access, and user approvals in the active harness. Jira and Slack MCP tool names and calls are retained as documented in the skills.

Generated reports

The Renovate and sprint-report skills write reports under .conforma/reports/ in the current project. Renovate's --report <path> option can still read and update an existing report at another path, including .claude/reports/; the skills do not move or delete old reports automatically.

Migration from conforma/claude-skills

The canonical repository is now conforma/agent-skills. Replace the old clone and install the complete directories from skills/ into your harness's skill location. Do not copy the old .claude/ tree from a prior checkout. If you have local reports under .claude/reports/, keep them and pass their JSON path to renovate-act --report <path> when needed.

The old slash-command entry points are now named skills. Provide their existing inputs in your request or use the invocation syntax supported by your harness. Their domain workflow and confirmation steps remain in each SKILL.md.

Contributing

  • Put reusable instructions in skills/<name>/SKILL.md; keep names lowercase and hyphenated, with a clear description.
  • Keep references, examples, templates, and scripts beside the skill that uses them. Link resources relatively to SKILL.md and do not rely on the caller's working directory.
  • State inputs, prerequisites, outputs, missing-input behavior, and failure paths.
  • Do not require model-specific features, client tool names, or harness settings in shared skill content. Keep client installation examples optional and outside the workflow instructions.
  • Preserve approval gates, trust boundaries, and MCP tool identifiers/calls when updating service workflows.
  • Run the local checks described in the pull request workflow before submitting changes.

License

Apache 2.0

Ver o original no GitHub

Bom para

Relacionados