Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add skills/firstintent/a2a-bridge/parallelnpx skills add firstintent/a2a-bridge --skill parallelgit clone --depth 1 https://github.com/firstintent/a2a-bridgeWhat it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.00043 | $0.01203 |
| Opus 5 | $0.00022 | $0.00602 |
| Sonnet 5 | $0.00009 | $0.00241 |
| Haiku 4.5 | $0.00004 | $0.00120 |
Grade A, and why
parallel scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured yesterday.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 133 lines — stays where its author put it; the contents beside it link to each section on GitHub.
parallel — spawn N independent peers, merge when they finish
The only safe multi-agent parallel pattern: carve the work into subtasks that do not depend on each other's outputs, send each to a peer, await all of them, then merge. If any subtask depends on another's result, this pattern degrades to sequential with extra steps and should not be used.
When to use
- Subtasks are independent by construction — different files, different modules, different investigations, different doc pages.
- Each subtask's output is stable regardless of the others' outcomes.
- Wall-clock latency matters enough to justify N× the token cost. (The equivalent sequential run costs 1× tokens; parallel costs N× and only saves time.)
When NOT to use
- Subtasks share context or build on each other's output. Merge loses fidelity; sequential with the real output is better.
- The "independence" is by job title (planner / implementer / tester / reviewer). This is sequential pretending to be parallel and almost always underperforms one session with the same tools.
- One of the subtasks would be cheap enough to inline. Parallel overhead only pays off when every branch is substantial.
Protocol
Spawn one message/stream call per subtask. Each call gets its own
contextId so replies do not cross wires; a2a-bridge's RoomRouter
(Phase 4) isolates them end-to-end. Then await every SSE stream to
its terminal status-update.final === true and merge.
┌─ subtask A ─────────► peer 1 ─► artifact-update A ─► final A ┐
├─ subtask B ─────────► peer 2 ─► artifact-update B ─► final B ┤─► merge
└─ subtask C ─────────► peer 3 ─► artifact-update C ─► final C ┘
return_format per branch is a caller choice:
"full"when each branch's raw output matters to the merge."summary"when only a conclusion per branch matters (seeskills/context-protect/)."verdict"when each branch is a check (seeskills/verify/); merge ispassonly when all branches returnpass.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- yesterday First seen · 133 lines · 43 tokens per session scan A 0a1af6d839d4
parallel is a skill published in the GitHub repository firstintent/a2a-bridge (8 stars, last pushed 4mo ago), licensed MIT. It adds 43 tokens to every session and 1,203 once invoked, about $0.0002 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other skills, from other repositories
qa-testing
Verify your work by actually operating the app or website you changed, instead of assuming it works. Strongly recommended whenever you build, modify, or debug a web app, website, or desktop GUI app. Drive real browsers with the agent-browser CLI and native desktop apps with the cua-driver CLI. These are installed on…
blog-writer
Peri 项目博客写作风格指南。当用户说"写博客"、"写文章"、"出稿"、 "按风格写"、"博客"时触发。也适用于用户丢过来素材说"帮我写篇博客"的场景。 覆盖项目介绍、技术复盘、架构讨论、性能优化、架构设计等类型。.
go-127
What changed in Go 1.27 (released August 2026) and how it changes the way Go is written in pi-go. Use this skill when writing or reviewing Go that could use a 1.27 feature, when bumping the go directive in go.mod, when a build or test behaves differently after a toolchain upgrade, or when code-guidelines-go points…
auto-devflow
Use when starting an issue, bugfix, feature, or refactor that benefits from an adaptive development workflow. Select lite, normal, pro, max, or ultra from task complexity and risk, then use only the coordination, review, and verification phases that the task actually needs.
langfuse
Interact with Langfuse and access its documentation. Use when needing to (1) query or modify Langfuse data programmatically via the CLI — traces, prompts, datasets, scores, sessions, and any other API resource, (2) look up Langfuse documentation, concepts, integration guides, or SDK usage, or (3) understand how any…
self-build
Builds isolated npm capability packages that operate on real project code and connects them to Peri through MCP/MCPP and MetaHarness. Use when adding tools, resources, remote skills or agents, creating a Bun/Node.js stdio server, linking .mcp.json, or changing the active prompt and middleware set.