Agent Skills

Portable Agent Skills for browser automation, contract decompilation, correctness and source security audits, Codex Goal checkpoint monitoring, OpenBao credential management, on-chain diagnosis, PocketBase schema design, provider-scoped model selection, safe Obsidian writes, and Vykar operations.

Skills

• browserless — Capture screenshots, render PDFs and automate browser sessions through a configured Browserless service.
• dedaub-decompile — Decompile deployed EVM contracts through a signed-in Dedaub account and save the source locally.
• goal-longpoll — Wait for exact Codex Goal checkpoints through a local MCP server, with optional session recovery hooks.
• gui-responsiveness — Diagnose and fix UI stalls using profiling, watchdogs and measured before-and-after results.
• model-effort-recommendation — Choose a model and reasoning setting using official guidance and verified availability in the invoking client.
• obsidian-safe-write — Safely update Obsidian vault files with storage-capacity checks and final-file verification.
• openbao-ops — Manage OpenBao credentials, scoped access policies and protected snapshots through a configured broker.
• pocketbase-schema — Design PocketBase schemas, authentication and permissions, and verify access boundaries with native fixtures.
• s3-ops — Manage S3-compatible buckets, objects, lifecycle rules and access checks through an explicitly configured endpoint.
• security-audit — Run dependency-remediation audits, application correctness reviews and on-chain diagnosis.
• strix-audit — Independently audit a frozen repository with Strix using DeepSeek V4.1 Flash, GLM 5.3 or GPT-6 Luna, honoring selected models and provider presets.
• vykar-ops — Install and update Vykar, configure backups, recover snapshots and verify backup integrity.

Supported clients

Client	Project location	User location
AGY / Google Antigravity	.agents/skills/<skill-name>/	~/.gemini/antigravity-cli/skills/<skill-name>/ (CLI); ~/.gemini/config/skills/<skill-name>/ (IDE)
Claude Code	.claude/skills/<skill-name>/	~/.claude/skills/<skill-name>/
Codex CLI	.agents/skills/<skill-name>/	~/.codex/skills/<skill-name>/
Cursor	.cursor/skills/<skill-name>/ or .agents/skills/<skill-name>/	~/.cursor/skills/<skill-name>/ or ~/.agents/skills/<skill-name>/
GitHub Copilot	.github/skills/<skill-name>/, .claude/skills/<skill-name>/, or .agents/skills/<skill-name>/	~/.copilot/skills/<skill-name>/ or ~/.agents/skills/<skill-name>/
OpenCode	.opencode/skills/<skill-name>/, .claude/skills/<skill-name>/, or .agents/skills/<skill-name>/	~/.config/opencode/skills/<skill-name>/, ~/.claude/skills/<skill-name>/, or ~/.agents/skills/<skill-name>/

Copy the complete skill directory to the applicable location. Each directory follows the open Agent Skills format and can also be installed in another client that explicitly documents support for that format.

Each agents/openai.yaml file is runtime metadata consumed by Codex. The other YAML files in agents/ are compatibility records with verified client paths and activation behavior; those clients consume SKILL.md directly rather than reading the sidecars.

goal-longpoll requires Python 3.11+ and read access to the current Codex Goals SQLite database. Its optional producer recovery requires psutil outside Linux. It is Codex-specific; Windows ACL enforcement is not implemented, so sensitive checkpoint data is unsupported there. openbao-ops requires Python 3.11+ and an existing HTTPS OpenBao service. Its broker supports Unix sockets on Linux/macOS and restricted named pipes on Windows. strix-audit requires Python 3.11+ and Git on Linux. DeepSeek and GLM use OpenRouter with a protected API credential and support Strix 1.6.2; selecting GPT requires Strix 1.7.0 and ChatGPT subscription sign-in; macOS is unverified and Windows is unsupported. Full audits also require Docker. Other skills retain their documented runtimes. Install only the skills and dependencies you need.

Keep endpoints, credentials, vault paths, runtime caches, and service-specific settings in the external configuration locations documented by each skill.

Each skill directory is an independent distribution and audit unit. Audit its supplied instructions, scripts, references, assets and declared dependencies together.

The security-audit dependency-remediation route requires a separately installed and configured ICU Security MCP server. This repository supplies the skill instructions, not that server or its scanners. The independent Strix phase also requires the strix-audit skill and its documented runtime.

Validate each skill against the Agent Skills specification after installation. Run a skill's focused tests when changing its helpers.

License

MIT — see LICENSE.