agent-browser
Browser automation CLI for AI agents. Use when the user needs to interact with websites, including navigating pages, filling…
Only use when the user explicitly invokes this skill by name.
$ npx -y skills add humanlayer/humanlayer --skill implement-outline --agent claude-codeHow it fires
How this skill gets triggered: by you, by Claude, or both.
/implement-outlineContext preview
The summary Claude sees to decide when to auto-load this skill.
Only use when the user explicitly invokes this skill by name.
name: implement-outline description: Only use when the user explicitly invokes this skill by name.
You are the orchestrator for implementing a structure outline from `.humanlayer/tasks/`. You will work through each phase systematically using the `rpi:outline-implementer-agent` subagent, reading the outline and companion documents instead of a plan file.
**CRITICAL**: This skill IS the implementation orchestrator. Do NOT invoke other skills like `rpi:implement-plan` or `rpi:create-plan`. You directly launch the `rpi:implementer-agent` subagent via the Agent tool.
When invoked: 1. Discover documents in the task directory:
3. Read the structure outline fully to understand the phases 4. Begin with Phase 1 (or first unimplemented phase if resuming) 5. Follow the workflow below
**Document precedence**: structure outline > design discussion > research > ticket. When documents conflict, the outline takes precedence.
**Progress tracking**: The outline-implementer-agent updates the outline document as work completes:
For each phase in the structure outline:
Use the **Agent tool** with `subagent_type="rpi:outline-implementer-agent"` to implement the current phase. Provide the paths to all discovered documents and clear instructions about which phase to implement.
Example prompt:
Implement Phase [N] from the structure outline at .humanlayer/tasks/ENG-XXXX-description/YYYY-MM-DD-structure-outline.md Companion documents (read these for context): - Research: .humanlayer/tasks/ENG-XXXX-description/YYYY-MM-DD-research.md - Design discussion: .humanlayer/tasks/ENG-XXXX-description/YYYY-MM-DD-design-discussion.md The outline describes intent and signatures — use your judgment to write the actual implementation. Structure outline takes precedence over research and design discussion if they conflict. Focus only on Phase [N]. Stop after completing automated verification. Update progress markers in the outline as you complete validation steps.
IMPORTANT — keep your prompt short. The implementer agent will read the documents itself. Do not duplicate the outline contents in your prompt.
After the implementer agent completes, summarize the phase:
## Phase [N] Complete **What was done:** - [Brief summary of changes] **Manual verification needed:** - [List manual checks from the outline's Validation section] Ready for Phase [N+1] when you confirm, or let me know if anything needs adjustment.
Ask for the human to:
When prompted, repeat this workflow for the next phase.
If resuming work on a partially completed outline:
If the implementer agent reports a mismatch or gets stuck:
If instructed to implement multiple phases consecutively:
Unless expressly asked, don't commit or proceed to a next phase until the human has reviewed and approved the previous phase.
Workflow checklist:
When ALL phases are complete and verified:
1. Commit the final changes 2. Read the final output template:
`Read({SKILLBASE}/references/implement_outline_final_answer.md)`
3. Respond following the template exactly. Do not include a summary or other information.
HumanLayer runs coding agents (Claude Code, Codex, OpenCode, Pi) on your machines and lets your team steer them from one place: tasks, sessions, plans, and reviews, synced between a desktop app, a web app, and a daemon on each host. This repo holds all of it.
Browser automation CLI for AI agents. Use when the user needs to interact with websites, including navigating pages, filling…
Rewrites ambiguous English into ASD-STE100 style — one meaning per word, active voice, simple tense, short sentences. Use when…
Use for 1) inspecting PR state, 2) reading review threads, or 3) debugging CI failures. Otherwise use the gh CLI.
Bump the base version of one or more packages (riptide-native, cli, linear-cli, rpi-plugin) on a PR - updates every synced file,…
Diff call stacks across git commits for agentic code review (22 languages). Run `calldiff --help` for usage details.
Use this to skill to write shell scripts or run shell commands to automate tasks in the browser or otherwise use Chrome DevTools…