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 skills add damianvtran/local-operator --skill verification-before-completiongit clone --depth 1 https://github.com/damianvtran/local-operatorWrote 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/skills/damianvtran/local-operator/verification-before-completion)<a href="https://agentmods.dev/skills/damianvtran/local-operator/verification-before-completion"><img src="https://agentmods.dev/badge/skills/damianvtran/local-operator/verification-before-completion/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/skills/damianvtran/local-operator/verification-before-completion"><img src="https://agentmods.dev/badge/skills/damianvtran/local-operator/verification-before-completion.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 | $0.00049 | $0.00665 |
| Opus 5.5 | $0.00020 | $0.00266 |
| Sonnet 5.5 | $0.00010 | $0.00133 |
| Haiku 4.5 | $0.00005 | $0.00067 |
Grade A, and why
verification-before-completion 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 5d 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 — 55 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Verification before completion
No claim ships without evidence you produced yourself, against the current state of the work.
Claims need evidence
| Claim | Counts as evidence | Does not count |
|---|---|---|
| "The fix works" | Rerun of the original reproduction, output shown | "Should work now" reasoning |
| "Tests pass" | The test command and its actual output | A remembered pass from before the change |
| "The endpoint works" | Real request, real response (status, body) | A unit test that mocks the call |
| "The UI shows X" | A rendered frame you captured and looked at | Reading the component source |
| "It fails safely" | The failure case exercised (unauthorized, empty, invalid) | Presence of a try/catch |
| "Deployed" | Success output plus a request against the new version | The commit that should have triggered it |
| "The subagent finished" | Its output read, its central claim spot-checked | The task marked complete |
Run it, don't say it
- Exercise the real path: the actual command, the actual endpoint, the actual page.
- Read the output. A non-zero exit or an error string is not success; read what came back.
- Freshness: evidence must come from the current state. A run from before the last edit proves nothing about the edit.
- One claim, one piece of evidence. Where you cite a number or status, derive it or name its source.
When you cannot verify
Say so, in this shape: verified: X (evidence); unverified: Y; reason: Z. Never let an unverified item pass silently as done. An explicit "unverified" is an honest result; an implied pass is not.
Verifying other people's reports
The same bar applies to summaries, subagent reports, and CI statuses:
- A "done" from another agent is a lead, not evidence. Spot-check the load-bearing claim, cheaply.
- A summary that cites its sources can be trusted one level deeper only where the source exists and says what the summary claims.
Red flags
- "It compiles" offered as proof of behaviour.
- Quoting a previous test run instead of running the suite now.
- Screenshots or output from before the change under review.
- A long list of "should" statements.
- Reporting a percentage, count, or status from memory.
- Tests passing treated as the feature working; a green suite is the code doing what the tests expect, and the tests may not cover the change.
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.
- 5d ago First seen · 55 lines · 49 tokens per session scan A 594368e434d3
verification-before-completion is a skill published in the GitHub repository damianvtran/local-operator (218 stars, last pushed today), licensed MIT. It adds 49 tokens to every session and 665 once invoked, about $0.0002 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 2026-10-11.
Other skills, from other repositories
agent-eval
Use when measuring whether an LLM or agent system actually got better and gating merges on it: golden sets, fixing an inflated LLM-as-judge, scoring RAG (faithfulness, contextual recall) or agent trajectories (tool correctness, completion), or picking an eval framework. NOT building the agent loop, tools or RAG…
building-agents
Use when building or restructuring an LLM agent — provider adapter, tool calling, structured output, RAG, agent loop, eval gate, cost routing, tracing, MCP server — model-agnostic across OpenAI/Anthropic/Gemini/OSS so a model swap is a config change. NOT vector-store SQL alone (that is postgresdb) or service…
media-validator
Validates media files (MP4, PNG, TXT) by running OCR, metadata checks, and pattern matching against expected content, returning PASS/FAIL with captured evidence. Use when a task needs to verify a recording, screenshot, or text-evidence file actually contains an expected pattern before a PASS is claimed — the…
skill-test
Verifies that the skill system is working correctly.
mcplab-assistant
Operator guide for MCPLab config authoring, Test Case Assistant workflows, execution, and result analysis. Use when users need to create or refine test cases from runs/traces, suggest deterministic checks or value capture, write or debug MCPLab eval YAML, run or queue evaluations, troubleshoot failures, or compare…
synthetic-remote-code-execution
Synthetic inert detector regression fixture.