Mcp Audit — independently scanned and version-tracked by SaferSkills.
SaferSkills independently audited Mcp Audit (Agent Skill) and scored it 91/100 (green). The audit ran 55 deterministic rules across Security, Supply Chain, Maintenance, Transparency, and Community; it found 1 high-severity and 0 lower-severity findings. The full rule-by-rule trace and per-finding evidence are below. Free, methodology-open.
Findings & checks · 1 flagged
The text {match} is the classic direct prompt-injection phrasing. Placed in a skill body that the agent reads as trusted instructions, it tries to make the agent abandon its prior rules and follow whatever comes next — a full system-prompt override.
ignore/disregard/forget … previous instructions sentence.Every scanned point with the score it earned and what moved between them.
First recorded scan — no prior version to compare against.
The primary manifest — the file an agent reads to learn what this artifact does.
Free-tier context engineering audits for production AI agents, delivered as MCP tools you can call from Claude Desktop, Cursor, Claude Code, or any MCP-compatible client.
What you get: top-3 findings on your system prompts, tool definitions, or context packing, on demand, no account needed.
What it costs: nothing. The free scan is genuinely free. Upgrade paths to the $49 Instant Audit and $750 Full Audit are surfaced in the response footer; they're not paywalls on this tool.
Most production agent failures aren't model failures — they're context engineering failures. Ambiguous instructions, underspecified tools, bloated context, no regression tests on prompt changes. Those problems are spottable by a trained reader. Archonics has trained that reader and published it as an MCP tool so you can get a second opinion on your agent's context without filing a support ticket.
The underlying audit engine applies Archonics Audit Methodology v1.0, the same spec that drives our paid audits.
audit_system_promptPaste a system prompt. Get back the three most important context engineering issues in it, ranked by severity, with specific recommendations.
Covers: role clarity, instruction conflicts, negative space, priority structure when instructions conflict, token efficiency, format specification precision, failure-mode coverage.
audit_tool_definitionPaste a tool/function definition. Get back the three most important issues affecting how reliably the model will call it.
Covers: description quality (the "when to use this tool" question), parameter schema precision, parameter documentation, error response design, discoverability.
audit_context_packingPaste a representative context payload (or describe it structurally). Get back the three most important efficiency and quality issues.
Covers: content inventory, redundancy across sections, freshness/relevance, ordering, truncation risk, prompt-cache utilization.
Add to your claude_desktop_config.json:
{
"mcpServers": {
"archonics-audit": {
"command": "npx",
"args": ["-y", "@archonics/mcp-audit"],
"env": {
"ANTHROPIC_API_KEY": "your-anthropic-api-key-here"
}
}
}
}Add to your .cursor/mcp.json:
{
"mcpServers": {
"archonics-audit": {
"command": "npx",
"args": ["-y", "@archonics/mcp-audit"],
"env": {
"ANTHROPIC_API_KEY": "your-anthropic-api-key-here"
}
}
}
}claude mcp add archonics-audit npx -y @archonics/mcp-auditThen set ANTHROPIC_API_KEY in your environment.
The audit engine runs on Claude. You bring your own API key so:
If you'd rather not bring your own key, use the $49 Instant Audit at agent.market — we cover the API costs and return a full-methodology audit PDF.
Submitted content is processed ephemerally. No prospect content is retained on Archonics infrastructure or used to train any model. The API call pattern is: your client → your Anthropic API key → Anthropic → your client. Archonics servers are not in this path.
Aggregated, anonymized patterns across many audits may inform improvements to the methodology — "18 of 20 audited systems lacked prompt-regression tests" — but specific content never feeds that process.
Details: archonics.ai/privacy
If the free scan surfaces issues worth fixing, two paid tiers go deeper:
MIT. Use it, fork it, audit yourself.
~30 seconds. Free. No account. Every finding cites a rule and a line of evidence.