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 Kevin-Liu-01/Agent-Machines --skill closed-loop-developmentgit clone --depth 1 https://github.com/Kevin-Liu-01/Agent-MachinesWrote 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/kevin-liu-01/agent-machines/closed-loop-development)<a href="https://agentmods.dev/skills/kevin-liu-01/agent-machines/closed-loop-development"><img src="https://agentmods.dev/badge/skills/kevin-liu-01/agent-machines/closed-loop-development.svg" alt="Measured on agentmods" height="20"></a>- NVIDIA SkillSpector warn
SkillSpector: 2 findings, up to medium
These are SkillSpector’s own severities. On a checked sample its high-severity flags on skills were ~96% false positives — a documented command, a public API, a “never do X” rule — so we show them as a caution to read, not a verdict. Why →
- medium Excessive Agency · line 3 Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.Fix: Add human-in-the-loop confirmation for destructive, irreversible, or high-impact operations. Never auto-execute commands that modify files, send data, or alter system state.
- medium Rogue Agent · line 3 Skill establishes unauthorized persistence across sessions via cron jobs, startup scripts, or state files. Session persistence allows an attacker to maintain access beyond the current interaction.Fix: Remove any persistence mechanisms (cron jobs, startup scripts, state files). Skills should not maintain state across sessions without explicit user consent.
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 | $0.00052 | $0.00619 |
| Opus 5 | $0.00026 | $0.00309 |
| Sonnet 5 | $0.00010 | $0.00124 |
| Haiku 4.5 | $0.00005 | $0.00062 |
Grade A, and why
closed-loop-development scanned grade A with 1 finding 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 8d 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.
Makes network callslowCapability
Not a fault in itself. Listed so you know the mod talks to something, and to what.
- API: use `curl`, `httpx`, and `jq`. Hit the actual route. Save the exact response shape before changing client code around it. How it starts
The opening of the file, as written. The whole thing — 46 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Closed-Loop Development
Agent Machines are built so I can verify my own work. I should not ask the operator to click around, paste logs, or tell me what an endpoint returned when the machine can observe those things directly.
Default loop
- Read the repo instructions and identify the narrow behavior to change.
- Make the smallest useful edit.
- Start the relevant server, worker, or test target.
- Exercise the real surface with the right tool.
- Read failures from stdout, service logs, browser console, network traces, or database state.
- Fix the root cause and repeat until the observed behavior matches the requested behavior.
Tool choices
- Browser/UI: use
agent-browseror thebrowser_*toolset to navigate, snapshot, interact, screenshot, and inspect rendered UI. Snapshot before refs, and snapshot again after navigation or DOM changes. - API: use
curl,httpx, andjq. Hit the actual route. Save the exact response shape before changing client code around it. - Database: use
sqlite3for local SQLite files and migration checks. Query the schema after running migrations. - Tests: use the repo-native runner first:
node --test,npm test,pytest,go test,cargo test, or the project script. - Logs: inspect
/.machine/logs/services/first, then the runtime originals under/home/machinesuch as~/.agent-machines/logs/gateway.log. - Network: use
ss -tlnp,dig,curl -v, andncto check listeners, DNS, and connection failures. - Runtime docs: read
/.agent/llm.txtand/.agent/docs/agent-context.mdbefore assuming which machine tools exist.
When to stop
Stop and report only when the blocker requires human ownership: missing credentials, captcha/passkey, destructive confirmation, provider outage, or an unavailable external system. Include what I tried, what I observed, and the next concrete manual step.
Anti-patterns
- Asking "can you check the console?" when service logs or browser console tools exist.
- Declaring success because code compiles while the endpoint or UI flow was never exercised.
- Adding retries or fallbacks before proving the root cause.
- Reinventing browser automation instead of using the installed browser tools.
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.
- 8d ago First seen · 46 lines · 52 tokens per session scan A bdcbafac4146
closed-loop-development is a skill published in the GitHub repository Kevin-Liu-01/Agent-Machines (26 stars, last pushed 7d ago), licensed MIT. It adds 52 tokens to every session and 619 once invoked, about $0.0003 per session on Opus 5. A static security scan graded it A with 1 finding (makes network calls). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other skills, from other repositories
aar-operations
Operate the Adaptive Agent Runtime through its public MCP tools. For tasks that need tool use or analysis, consider AAR MCP early when bounded stateful computation, brokered jobs, operations, contracts, or artifacts can materially help; software planning, development, testing, and troubleshooting belong to this class…
systematic-debugging
4-phase root cause debugging: understand bugs before fixing.
experimental-code-coverage-local-debugger
Runs code coverage locally via Universal Test Runner (UTR) or helper scripts, mimicking LUCI trybots. Activate when CQ tryjobs fail or underreport coverage, to test local GN/recipe repairs before uploading, or to debug hermetic crashes.
ios-simulator
Verify and debug native, React Native, Expo, or Flutter apps on an iOS Simulator with agent-device. Use when an agent needs to launch an app, inspect its live UI, tap, type, scroll, validate a code change, collect failure evidence, or reproduce a workflow on an iPhone or iPad Simulator.
systematic-debugging
Use when encountering any bug, test failure, or unexpected behavior, before proposing fixes.
test-first-bugs
Enforces a test-driven bug-fixing workflow. Use when a user reports a bug, failing code, an error, or asks to fix something.