Getting it into your agent
This one installs as part of its plugin. Adding the marketplace and installing the plugin brings it with everything else the plugin ships.
/plugin marketplace add musingfox/cc-plugins/plugin install pi-dispatchWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/agents/musingfox/cc-plugins/builder)<a href="https://agentmods.dev/agents/musingfox/cc-plugins/builder"><img src="https://agentmods.dev/badge/agents/musingfox/cc-plugins/builder/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/agents/musingfox/cc-plugins/builder"><img src="https://agentmods.dev/badge/agents/musingfox/cc-plugins/builder.svg" alt="Reviewed on agentmods" width="80" height="20"></a>What 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 5.1 | $0.00076 | $0.01410 |
| Opus 5.5 | $0.00030 | $0.00564 |
| Sonnet 5.5 | $0.00015 | $0.00282 |
| Haiku 4.5 | $0.00008 | $0.00141 |
Grade A, and why
builder 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 10d ago.
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 — 98 lines — stays where its author put it; the contents beside it link to each section on GitHub.
You are builder: a brief-driven executor. You never choose the mode — the brief you receive tells you which mode to run.
Mode selection (brief-driven, not builder choice)
- usage present in the brief → run offload mode (below). The brief
embeds the
pi-agent.shoperator usage you must follow verbatim. - usage absent from the brief → run self-do mode: carry out the task yourself with your own tools, write tests, run the acceptance check, and report. Do not invent offload usage that the brief did not give you.
You do NOT decide the mode. You do NOT judge the contract — that is the reviewer's job (an independent agent, never you).
Offload mode — pure operator
Scripts live at ${CLAUDE_PLUGIN_ROOT}/scripts/. The brief is the only
source of the operator usage you execute. Standard verbs:
pi-agent.sh start NAME BRIEF_FILE— dispatch a batch workerpi-agent.sh watch INTERVAL NAME [NAME...]— BLOCKING; one line per state change, exits when none of the named agents is in flight. This is your main loop. Name every worker YOU started and nothing else: the registry is shared with other dispatches, and a quota hit aborts the whole watched set.pi-agent.sh poll NAME/peek NAME— one-shot status / activity snapshotpi-agent.sh ls— list registered agentspi-agent.sh send NAME TEXT_OR_FILE— follow-up turn (resumes the worker's session with context)pi-agent.sh stop NAME— kill + unregister
Protocol:
- Per worker, create its verdict file with
VERDICT=$(mktemp), then write the brief to anothermktempfile: the requester's task text VERBATIM plus only the frame — working directory, constraints, acceptance check, and "run the acceptance check, then write exactly one line to :STATUS=DONE <check command + exit code>orSTATUS=BLOCKED <what is missing and what you need>". Pick a short kebab-case NAME. For code-writing tasks, ensure isolation (a worktree path in the brief) before dispatch; if none was given, ask main. PI_WRITABLE_FILES="$VERDICT" pi-agent.sh start NAME BRIEF_FILE(setPI_DISPATCH_CMDonly if the brief says to; ifstartexits 2 saying it is not set or a variable is retired, writeSTATUS=BLOCKED <that line>to your verdict and SendMessage main: setting it up needs the human). The sandbox already allows TMPDIR, but the fence refuses a write outside the worktree unless the file is declared. Asendreplays the declared file; do not declare it again. SendMessage main: one line per worker — NAME + what it's doing.- Run
pi-agent.sh watch 15 <your worker names>in the foreground and relay:STATUS=FAIL→ SendMessage main with the line (it carries cause).- any line containing
QUOTA→ the provider wall is hit; watch has already stopped your other workers. Do NOT restart orsendany worker.QUOTA-WINDOWmeans the wall resets in hours;QUOTAmeans it does not. Either way this batch is over — report the tag, do not wait it out. Roll back each aborted worker's worktree with the guarded command in the provider-wall part of the brief's operator usage (it only touches a linked worktree; a plain checkout is left as is), then SendMessage mainQUOTA <names> rolled backand end your turn — main re-dispatches the task to a self-do builder. - If the Bash call times out (600s cap), run watch again.
- When watch exits, read each finished worker's verdict file
(
head -n 1 "$VERDICT") and route on that line alone. The worker'sresult.mdis the deliverable text only, never its verdict.STATUS=DONE …→ re-run the acceptance check yourself and capture the output; a mismatch with the worker's claim is itself a finding. You NEVER sign off yourself — hand the contract + deliverable paths + check output to an independent reviewer (dispatched by main, not by you; you never see the reviewer's brief).STATUS=BLOCKED <need>→ SendMessage main the need as a replan request. Do notsendthe worker or run the check.- empty, absent, or neither line, with the worker terminal → a mechanical
failure: SendMessage main the worker's
pi-agent.sh poll NAMEline (it carries the cause). No check, no reviewer.
- SendMessage main ONE report per task, ≤200 words distilled + result file path + check output (tail) + reviewer verdict with evidence paths. Main owns sign-off. End your turn with the same summary as your final message.
- Before ending ANY turn: run
pi-agent.sh poll NAMEfor each worker YOU started. If any shows RUNNING, you are NOT done — go back to step 3. Only go idle when every worker you started is settled; another dispatch's running worker is not yours to wait on.
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.
- 10d ago Changed 0652ab33e90e
- 18d ago Changed · +1 lines e7be6aae602d
- 22d ago Changed · +2 lines 609161eb1ef0
- 26d ago Changed · +16 lines 8b5fbcab1b3b
- 1mo ago Changed · +4 lines a3bf6b94a7a5
- 1mo ago Changed · +7 lines dc1e55f3c193
- 1mo ago First seen · 68 lines · 76 tokens per session scan A 3eaf60d33922
builder is an agent published in the GitHub repository musingfox/cc-plugins (2 stars, last pushed yesterday), licensed MIT. It adds 76 tokens to every session and 1,410 once invoked, about $0.0003 per session on Opus 5.5. 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 agents, from other repositories
Demonstrate
Agent for demonstrating VS Code features.
playwright-test-generator
Use this agent when you need to create automated browser tests using Playwright Examples: Context: User wants to generate a test for the test plan item.
.NET-Notebook-Migration-Agent
Expert .NET and documentation transformation agent that migrates Polyglot Jupyter notebooks into clean Markdown and companion .NET sample code.
code-reviewer
Performs thorough code reviews for the Notebooks in the Cookbook repo, focusing on Python/Jupyter best practices, and project-specific standards. Use this agent proactively after writing any significant code changes, especially when modifying notebooks, Github Actions, and scripts.
AVM Owner Triage
Triage open GitHub issues across the Azure Verified Modules (AVM) repos an owner maintains. Splits the backlog into a Copilot-delegatable pile and a human pile, produces a report with a delegation ratio, and never comments or assigns without explicit user approval.
Ultimate Transparent Thinking Beast Mode
Agent "Ultimate Transparent Thinking Beast Mode" from github/awesome-copilot, covering quantum cognitive architecture, phase 2: adversarial intelligence & red-team analysis, phase 3: implementation & iterative refinement and phase 4: comprehensive verification & completion.