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/sma1lboy/rove/verifying-worker-outputgit clone --depth 1 https://github.com/Sma1lboy/roveWrote 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/sma1lboy/rove/verifying-worker-output)<a href="https://agentmods.dev/agents/sma1lboy/rove/verifying-worker-output"><img src="https://agentmods.dev/badge/agents/sma1lboy/rove/verifying-worker-output.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 | $0.00000 | $0.01513 |
| Opus 5 | $0.00000 | $0.00757 |
| Sonnet 5 | $0.00000 | $0.00303 |
| Haiku 4.5 | $0.00000 | $0.00151 |
Grade A, and why
verifying-worker-output 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 — 78 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Verifying Worker Output
What a coordinator does after a worker reports succeeded. Dispatching workers and the report contract they follow live in the rove skill (.claude/skills/kobe/SKILL.md); this page starts where that ends — succeeded is a claim, not a verification. Every rule here was paid for in one night of verifying 26 worker PRs; the examples are real.
flowchart TD
R["worker reports succeeded"] --> B{branch exists?}
B -- no --> X["report is unreliable —\nre-verify every claim in it"]
B -- yes --> C{commits ahead of base?}
C -- no --> E["EMPTY_BRANCH —\nnothing was committed"]
C -- yes --> D{diff contains only the work?}
D -- no --> P["cherry-pick onto a\nclean worktree"]
D -- yes --> T["prove the test can fail"]
P --> T
T --> CI["triage red CI:\nregression / flake / hang / stale green"]
CI --> L["land or open PR,\nwith honest caveats"]
First: does the branch exist, and does it hold commits?
A worker report can be wrong at the most basic level. One task sent two contradictory reports; one of them named a branch that did not exist, and its conclusion was the opposite of what the real branch contained. So before reading anything else in the report:
git rev-parse --verify <branch> # the branch is real
git log origin/main..<branch> --oneline # it is ahead of the base
Empty git log output means nothing was committed — the worker may have done the work and reported success with it still sitting uncommitted in the worktree. rove api land refuses this case with EMPTY_BRANCH (or EMPTY_BRANCH_DIRTY_WORKTREE when the uncommitted work is recoverable), so land failing this way is the signal, not an obstacle.
rove api collect --task-ids <a,b,c> batches this for a fleet: per task it returns the branch, uncommitted .changes, and .base (commits ahead + diffstat vs the base branch) in one read.
If the branch doesn't exist or is empty, stop trusting the report wholesale — everything else in it needs independent verification too.
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 · 78 lines · 0 tokens per session scan A 8f6df951733e
verifying-worker-output is an agent published in the GitHub repository Sma1lboy/rove (116 stars, last pushed today), licensed MIT. It costs nothing until one of its globs matches a file; then it loads 1,513 tokens. 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 2026-09-03.
Other agents, from other repositories
plan-chunks-agent
Use this agent for autonomous story planning — deep codebase research + detailed chunk breakdown in one focused pass. Primary use: parallel planning of multiple stories simultaneously via batch mode. Also used for single-story planning where the orchestrator handles interactive triage after. The orchestrator MUST…
implementer
Use this agent when implementing story chunks, building features from specs, or continuing active story implementation. Owns the implement → validate → refine loop for each chunk. Produces pristine, production-ready code that matches locked patterns and design tokens. Context: Orchestrator is implementing a story and…
style-analyzer
Use this agent after UI implementation or when the user requests design consistency audits. Ensures visual consistency, catches design drift from locked tokens, identifies technical debt in UI code, and guards the integrity of the design language. Context: Multiple UI components were built during the cycle. user…
tester
Use this agent after chunk implementation to create comprehensive test suites, or when the user requests test generation. Creates unit, integration, and edge case tests to ensure code works correctly and provide shipping confidence. Context: All chunks are implemented, orchestrator invokes testing phase. user: "All…
product-anthropologist
The human-truth layer for product decisions. Consult when diagnosing why users aren't adopting, when deciding whether to iterate or kill, when interpreting user feedback or metrics, when designing research for AI-powered products, when a founder's conviction is outrunning evidence, or any moment where the question is…
bug-reproducer
只写复现测试, 禁碰实现 —— 交付物是一条现在必须红的测试 (verify-red 判据).