Reusable agent skills maintained by Blaine Yeager. The repository is tool-agnostic: each skill documents its supported runtime and keeps its own scripts and references under skills/<skill-name>/.
Save paused work to a private local Markdown vault with an outcome, completion proof, next action, validity conditions, context, research spine, tools, and durable links. The included dependency-free CLI can list saved work and print a cold-resume brief with freshness warnings. It does not make network or external-service writes.
Install the standalone skill:
npx skills add https://github.com/blaineyeager/skills --skill park-handoffOr install the Codex plugin from this repository's marketplace:
codex plugin marketplace add blaineyeager/skills
codex plugin add park-handoff@blaineyeager-skillsThen ask: Park this work so another agent can resume it cold.
Run bounded Claude Code research jobs with a fresh session, isolated defaults, explicit tools, live JSON events, and turn and budget limits. The skill distinguishes observable claude -p work from tasks that require the Claude Agent SDK for live steering.
Install it with a compatible skills client:
npx skills add https://github.com/blaineyeager/skills --skill claude-headless-researchOr copy skills/claude-headless-research into the skills directory supported by your agent runtime.
skills/
├── claude-headless-research/
│ ├── SKILL.md
│ ├── agents/
│ ├── references/
│ └── scripts/
└── park-handoff/
├── SKILL.md
├── agents/
├── references/
└── scripts/
Review every skill before installing it. Scripts inherit the permissions of the calling process. Park Handoff writes only to its configured local vault, accepts private content through stdin, creates private-permission files where the platform supports them, scans persisted text for common credentials, and makes no network calls. Claude Headless Research requires a dedicated Claude config directory, ignores user and project settings, exposes read-only built-ins, and supplies no MCP servers by default. MCP-backed work requires a reviewed strict configuration. Organization-managed Claude policy has higher precedence than command-line isolation, so every run must reject unexpected additions reported by system/init.
MIT. See LICENSE.