Agent Claude Code
Read repo-root AGENTS.md first, then docs/backend-architecture.md, then LEARNINGS.md, before planning or implementing.
Agent Claude Code
Read repo-root AGENTS.md first, then docs/backend-architecture.md, then LEARNINGS.md, before planning or implementing.
Agent Claude Code
Implements features and fixes from a HANDOFF:PLAN. Writes and edits code, runs focused local checks, keeps scope minimal. Tests-first for any domain code under internal/ /. Hands off to smoke-tester via HANDOFF:IMPLEMENTATION.
Agent Claude Code
Creates GitHub issues, labels, and milestones using gh CLI. Breaks down features into runnable, well-labelled issues with binary acceptance criteria and local verification steps. Standalone — does not delegate to planner or any other agent unless explicitly asked.
Agent Claude Code
Picks the next GitHub issue from the current milestone, creates the linked task branch, explores the codebase, and writes a detailed technical plan. Analysis and planning only — no source edits, no commits. Delegates to the coder via HANDOFF:PLAN.
Agent Claude Code
Walks every assumption in HANDOFF:PLAN and tries to refute it by reading the repo. Upholds the plan (handing off to coder) or refutes it (escalating to user). Read-only — no code edits, no commits.
Agent Claude Code
Final review of the diff for correctness, security, conventions, and spec conformance. Waits for CI green. On clean verdict, appends one line to LEARNINGS.md and signals merge-ready. On blocking findings, emits HANDOFF:FIX to coder. Read-only on source code.
Agent Claude Code
Runs smoke tests after implementation — build, unit tests, start server if applicable, curl live endpoints, report results. Observer/recorder only — does not modify anything. Hands off to reviewer on pass, back to coder on fail.
Command Claude Code
Run the full delivery pipeline (planner → red-team → coder → smoke-tester → reviewer) against a GitHub issue. Bridges Claude Code's missing auto-handoff button by invoking each subagent in sequence via the Agent tool, validating typed HANDOFF blocks, breaking on repeat failures, and logging every stage to…
Settings file Claude Code
Agent settings declaring 8 allowed tools.
Cursor rule Cursor
How to use bastion subagents, HANDOFF blocks, and gh CLI in Agent mode.
Skill Claude CodeCodexCursor
Creates and structures GitHub issues and labels using gh CLI conventions. Use when creating or editing GitHub issues, triaging, defining label taxonomy, or running gh issue or gh label commands. Use when the user asks to create issues, plan a feature with issues, break down work into GitHub tasks, or set up labels and…
Agent
Implement a plan from IssuePlanner, run Bastion formatters and linters, commit, and open a PR.
Agent
Fetch GitHub issues, research the Bastion codebase, clarify with you, and produce a detailed implementation plan before handing off to IssueCoder. Use at the start of milestone work or when you need a plan before coding.
Agent
Walks every assumption in HANDOFF:PLAN and tries to refute it by reading the repo. Upholds the plan (handing off to Coder) or refutes it (escalating to user). Read-only — no code edits, no commits.
Agent
Read-only reviewer. Reads the PR diff, identifies issues, and hands off to IssueCoder to fix them. Never writes code.
Agent
Run smoke tests after implementation — build, unit tests, start server if applicable, curl live endpoints, report results.
MCP server Claude CodeCodexCursor +2
Lets the agent drive a real browser: open pages, click, type, take screenshots and read the accessibility tree, using Playwright. Runs locally from the @playwright/mcp npm package.
Instructions file CodexOpenCode
Instructions for JoakimCarlsson/bastion, covering agents.md — bastion contributor and agent contract, project summary, hard rules — verification (non-negotiable), smoke example and or: docker compose up --build.