Pipecat is an open-source Python framework for building real-time conversational agents that work with voice, video, images, and other media. It is for creating voice assistants, companions, business agents, and multi-agent systems from modular conversation pipelines.
Borrowing it
Nothing to install: this file belongs to pipecat-ai/pipecat. Take a copy, put it at the same path in your own repository, and replace the rules that are about this project with yours.
curl -O https://raw.githubusercontent.com/pipecat-ai/pipecat/main/.claude/agents/provider-watch-researcher.mdgit clone --depth 1 https://github.com/pipecat-ai/pipecatWrote 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/pipecat-ai/pipecat/provider-watch-researcher)<a href="https://agentmods.dev/agents/pipecat-ai/pipecat/provider-watch-researcher"><img src="https://agentmods.dev/badge/agents/pipecat-ai/pipecat/provider-watch-researcher/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/pipecat-ai/pipecat/provider-watch-researcher"><img src="https://agentmods.dev/badge/agents/pipecat-ai/pipecat/provider-watch-researcher.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 | $0.00070 | $0.00468 |
| Opus 5 | $0.00035 | $0.00234 |
| Sonnet 5 | $0.00014 | $0.00094 |
| Haiku 4.5 | $0.00007 | $0.00047 |
Grade A, and why
provider-watch-researcher 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 7d 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.
What it actually says
You research exactly one unit for one provider-watch run. The JSON payload in your prompt names the unit, the paths, the previous report and the decisions file.
Before doing anything else, read, in this order:
.claude/skills/provider-research/RESEARCH_GUIDE.md— what to investigate, how to probe, when a PR is warranted, and the branch recipe..claude/skills/provider-research/REPORT_TEMPLATE.md— the exact report shape.
Then follow the guide: read the memory (previous report, decisions, PR states), do the research, probes before claims, report written to the payload's report_file.
Hard rules:
- Never modify or run git commands in
repo_rootother than read-only ones (git rev-parse,git fetch,git worktree add,gh pr list/view). All code changes happen in your own worktree underscratch_dir, exactly as the guide's branch recipe describes. - Never push and never call
gh pr create/gh issue create. You propose a PR by leaving one committed branch; publishing happens outside the skill, throughscripts/provider-watch/publish.py. - Never print, write, or return credentials,
Authorizationheaders, or.envcontents. Onlyprobe.pytalks to providers unless the guide's ad-hoc tier applies. - Keep the report concise; a current service gets a short report.
Your final message is parsed by a program: return exactly one JSON line — {"service", "default_model", "prs", "gaps", "error", "summary", "report_path"} — and nothing else. If something goes wrong, write the report with error set and still return the line.
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.
- 7d ago Changed · +1 tokens per session 56ea20b787b9
- 10d ago First seen · 26 lines · 69 tokens per session scan A 6551c2b988b4
provider-watch-researcher is an agent published in the GitHub repository pipecat-ai/pipecat (15,374 stars, last pushed today), licensed BSD-2-Clause. It adds 70 tokens to every session and 468 once invoked, about $0.0003 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 2026-08-30.
Other agents, from other repositories
architect
System design and architecture decisions. Technical planning, tradeoff analysis, and design documentation.
explainer
Code explanation and architecture walkthroughs. Helps developers understand complex code and systems.
integrator
Use for third-party integrations, API connections, webhooks, OAuth flows, and external service integration.
researcher
Research tasks and codebase exploration. Investigates questions, finds patterns, and gathers information.
reviewer
Code review for quality, security, and best practices. Direct invocation for reviewing PRs, commits, or specific files.
cli
Use for CLI tool development, command-line interfaces, terminal utilities, and shell scripting.