From orca-strait
This skill should be used when the user asks to "run TDD agents", "orchestrate crate implementations", "spawn parallel test agents", "implement the plan with sub-agents", "run the handoff", "work through the issues", "implement open tasks in parallel", or any phrasing that implies reading issues/handoffs and dispatching per-crate sub-agents that write failing tests first. Also triggers on "/orca-strait".
How this skill is triggered — by the user, by Claude, or both
Slash command
/orca-strait:orca-straitThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Parallel TDD sub-agent orchestrator for Rust workspaces. Read context sources (GitHub
Parallel TDD sub-agent orchestrator for Rust workspaces. Read context sources (GitHub issues, handoff files, implementation plans), decompose work by crate, spawn independent agents with strict test-first discipline, then integrate the results.
The orchestration flow has four stages:
.ctx/, read all HANDOFF.* files, read the
implementation plan.mkdir -p .ctx/git/issues
gh issue list --state open --json number,title,body,labels,assignees \
> .ctx/git/issues/open.json
Parse open.json for task extraction in Stage 2. The file is the canonical source of
truth — do not re-query GitHub during decomposition.
# Glob pattern — picks up HANDOFF.yaml, HANDOFF.md, HANDOFF.json, etc.
ls HANDOFF.* 2>/dev/null
Read every file matching HANDOFF.* in the repo root. These are written by other agents
and contain current state, blocked items, and next steps.
Look for these files in order (first found wins):
PLAN.mdIMPLEMENTATION.md.ctx/plan.md*PLAN*.md in repo rootIf none exist, derive the plan from issues + handoff content.
Parse all context into a task list. For each task:
-p <crate> argument for cargo nextest).Independent tasks at the same crate level can run in parallel. Tasks that share a crate must run sequentially within that crate's agent.
Use the task decomposition template:
helpers/decompose.sh — prints a JSON task list from .ctx/git/issues/open.json
See templates/task-list.json for the schema.
Spawn one Agent tool call per independent crate, up to 5 concurrent. Each agent
receives this exact instruction set (see templates/agent-prompt.md for the full
template):
You are a TDD implementation agent for crate: <CRATE_NAME>
Tasks assigned: <TASK_LIST>
Workflow (follow exactly — do not skip steps):
1. Read only the source files in crates/<CRATE_NAME>/ relevant to the tasks.
2. For each task:
a. Write a FAILING test that captures the requirement.
b. Run: cargo nextest run -p <CRATE_NAME> -- <test_name>
Confirm it fails. If it passes already, the task may already be done — verify.
c. Implement the minimum code to make the test pass.
d. Run: cargo nextest run -p <CRATE_NAME>
All tests must pass before moving to the next task.
e. Run: cargo clippy -p <CRATE_NAME> -- -D warnings
Fix every warning. Commit only when clippy is clean.
3. If stuck after 3 attempts on any single task:
a. Write BLOCKED.md at the repo root describing what was attempted and why it failed.
b. Stop. Do not proceed to other tasks.
4. After all tasks pass: commit with message "feat(<CRATE_NAME>): <summary>".
Enforce the 5-agent cap strictly. If there are more than 5 independent crates, queue the remainder and dispatch each as a running slot frees up.
Reference: helpers/dispatch.sh — wraps the Agent dispatch loop with slot tracking.
After all agents report completion:
git log --oneline -5 per crate branch, or check
worktrees if using isolation: worktree).cargo nextest run --workspace
cargo clippy --workspace -- -D warnings and fix any remaining warnings.--allowedTools when
dispatching. Minimum set: Read,Write,Bash,Grep,Glob.| Step | Command |
|---|---|
| Sync issues | gh issue list --state open --json ... > .ctx/git/issues/open.json |
| List handoffs | ls HANDOFF.* |
| Run crate tests | cargo nextest run -p <crate> |
| Run workspace | cargo nextest run --workspace |
| Clippy strict | cargo clippy --workspace -- -D warnings |
| Check commits | git log --oneline -5 |
references/context-sources.md — Detailed schema for issues JSON, handoff YAML,
and plan file formats.references/agent-guardrails.md — Full list of sub-agent restrictions, tool
allowlists, and failure modes.references/tdd-discipline.md — Red/green/refactor cycle, when to skip steps
(almost never), and common TDD anti-patterns in Rust.helpers/decompose.sh — Parse issues JSON into a structured task list.helpers/dispatch.sh — Dispatch loop with 5-agent concurrency cap.templates/agent-prompt.md — Full sub-agent instruction template (copy verbatim).templates/task-list.json — JSON schema for the decomposed task list.npx claudepluginhub 89jobrien/bazaar --plugin orca-straitProvides UI/UX resources: 50+ styles, color palettes, font pairings, guidelines, charts for web/mobile across React, Next.js, Vue, Svelte, Tailwind, React Native, Flutter. Aids planning, building, reviewing interfaces.
Fetches up-to-date documentation from Context7 for libraries and frameworks like React, Next.js, Prisma. Use for setup questions, API references, and code examples.