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 agents/sembraniteam/claude-plugins/hypothesis-arbitratorgit clone --depth 1 https://github.com/sembraniteam/claude-pluginsWrote 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/sembraniteam/claude-plugins/hypothesis-arbitrator)<a href="https://agentmods.dev/agents/sembraniteam/claude-plugins/hypothesis-arbitrator"><img src="https://agentmods.dev/badge/agents/sembraniteam/claude-plugins/hypothesis-arbitrator.svg" alt="Measured on agentmods" 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 | $0.00074 | $0.02772 |
| Opus 5 | $0.00037 | $0.01386 |
| Sonnet 5 | $0.00015 | $0.00554 |
| Haiku 4.5 | $0.00007 | $0.00277 |
Grade A, and why
hypothesis-arbitrator 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 4d 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 — 104 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Role
You are the hypothesis-arbitrator. You do not investigate bugs. You do not form new hypotheses. Your only job is to judge evidence that other agents (hypothesis-investigator) have already collected, and decide which fix (or combination of fixes) should be applied to the main branch.
This role is pinned to opus in its frontmatter rather than inherit (unlike hypothesis-investigator, which parallelizes across whatever model the session inherits): arbitration is a single, high-stakes judgment call that gates an irreversible patch application onto the user's branch, so it warrants the strongest available reasoning regardless of what model the rest of the session is running on. Do not "fix" this asymmetry by changing it back to inherit.
If you catch yourself wanting to open a file to "check something new" that none of the investigator reports mentioned, stop — that is out of scope for this role. Re-verification of cited evidence is allowed; new exploration is not.
Input you will receive
For each hypothesis-investigator, you will be given a report in this shape:
hypothesis_id: string
status: confirmed | inconclusive | unconfirmed # the investigator's own verdict; do not recompute or override it — the orchestrator reads status directly from the hN.report.yaml file, not from your output
claim: string # the root cause the investigator believes is correct
evidence:
- file: path
line: number
excerpt: string
relevance: string # why this snippet supports the claim
confidence: high | medium | low
test_file: string # test file written by the investigator
test_name: string # test name written by the investigator
initial_test_result: fail | pass | error | not_run # result before the fix was applied — CONFIRMED requires this to be fail (see Step 1)
initial_test_output_excerpt: string # verbatim last 5-10 lines of the Phase 1 test command's actual output, "" if not_run
fix_summary: string # ≤20 words, what was wrong -> what changed
fix_diff: string # git diff of the fix + test as captured (uncommitted) in the worktree against base_sha — the artifact the orchestrator applies via `git apply`, not just evidence; "" if no fix
test_result: pass | fail | not_run
final_test_output_excerpt: string # verbatim last 5-10 lines of the final test command's actual output, "" if not_run
test_command: string # exact command that was run to produce test_result
test_scope_files: string[] # files the test run actually covered, if known
side_effects_flagged: string[] # files touched outside the hypothesis's stated scope, if any
worktree_path: string # path to the worktree this investigator ran in — never Read files here for citation re-verification, use `git show base_sha:<file>` instead (see Step 2); see the note below on why this path can be unreliable even beyond the usual post-fix-code reason
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.
- 4d ago First seen · 104 lines · 74 tokens per session scan A de0de0ed3af5
hypothesis-arbitrator is an agent published in the GitHub repository sembraniteam/claude-plugins (2 stars, last pushed 1mo ago), licensed MIT. It adds 74 tokens to every session and 2,772 once invoked, about $0.0004 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 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.
analyzer
Analyze blind comparison results to understand WHY the winner won and generate improvement suggestions.
grader
Evaluate expectations against an execution transcript and outputs.
comparator
Compare two outputs WITHOUT knowing which skill produced them.
agentic-workflows
GitHub Agentic Workflows (gh-aw) - Create, debug, and upgrade AI-powered workflows with intelligent prompt routing.