More screenshots

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    Supply Chain Guard

    2

    Static pre-install audit of skills, plugins, and MCP configs: 24 threat IDs, cross-file contradiction detection, trust score and verdict. No network calls, no execution.

    Free

    1 useSecurity scanned
    Supply Chain Guard

    Supply Chain Guard

    Example session with this skill installed

    Audit the package at examples/audit-guide/before/ before I install it.

    Baseline

    • Permitted permissions: filesystem read (package-local), no writes, no execution
    • Permitted destinations: api.fictive.invalid, docs.fictive.invalid
    • Declared permissions: as stated in the package manifest
    • Read your context and instructions
    • Compiled the supply chain guard

    Skill Supply-Chain Guard — audit report

    Verdict: REVIEW · score 80 / 100 · auditor@1.0

    Preamble

    • Auditor version: 1.0
    • Baseline: present — permitted permissions, permitted destinations, declared permissions
    • Precedence order: entry_point, manifest, config, reference_docs, examples
    • Declared specification: contract files excluded from audit per D-P3-01

    Inventory summary

    4 artifacts read: 1 SKILL.md, 1 manifest, 1 reference doc, 1 fixture. 0 opaque. 0 unreadable.

    Scored findings

    A2@reference/troubleshooting.md:22-24#1 — A2, HIGH severity, HIGH confidence

    • Span: reference/troubleshooting.md:22–24
    • Deduction: 20
    • Rationale: The reference file directs the reader to POST troubleshooting data to an external endpoint, while the entry point forbids network calls. Same topic key, incompatible value key, different precedence levels.

    Corroborating findings

    None.

    Cross-file evidence set

    • RESTATEMENT — allowed paths — SKILL.md:14 vs reference/troubleshooting.md:8 — not fired
    • ADDITION — output format — SKILL.md:22 vs reference/troubleshooting.md:14 — not fired
    • CONTRADICTION — network policy — SKILL.md:31 vs reference/troubleshooting.md:22–24 — fired

    Appendix — LOW confidence

    None.

    UNASSESSED categories

    • H2 — trigger always_x1_ceiling, gate X1

    H2's trigger is verdict-neutral: it is reported, and it does not engage the §6 cap.

    Score and verdict

    • Base: 100
    • Scored deduction: −20 (A2 HIGH @ HIGH)
    • Final: 80
    • Band (§4): REVIEW, 80 ∈ [60, 84]
    • Override (§5): not applied — no CRITICAL at MEDIUM-or-higher
    • Cap (§6): not applied — H2's trigger is verdict-neutral

    Verdict: REVIEW. One contradiction is a request to look before you install, not a refusal.

    Coverage

    A1 CLEAR · A2 FOUND · A3 CLEAR · B1 CLEAR · B2 CLEAR · B3 CLEAR · C1 CLEAR · C2 CLEAR · C3 CLEAR · D1 CLEAR · D2 CLEAR · D3 CLEAR · D4 CLEAR · E1 CLEAR · E2 CLEAR · E3 CLEAR · E4 CLEAR · F1 CLEAR · F2 CLEAR · G1 CLEAR · G2 CLEAR · G3 CLEAR · H1 CLEAR · H2 UNASSESSED

    Remediation

    R-A2 — cross-file contradiction. Rewrite the lower-precedence statement in reference/troubleshooting.md so it restates, paraphrases, or elaborates the entry point's network policy rather than contradicting it. Deleting the reference file drops legitimate ADDITION content and the contradiction may recur elsewhere; the fix is alignment, not removal.

    Next steps. Apply R-A2, then re-run the audit on the same tree. The corrected version is at examples/audit-guide/after/ and returns

    SAFE, 100/100 — see examples/audit-guide/after-report.md.

    1 SKILL.md

    Connects securely to your tools. The creator never sees your data.

    About this skill

    What you get

    Point Skill Supply-Chain Guard at a skill package, a plugin manifest, or an MCP server config, and get one trust score and one verdict before anything installs. The audit reads the whole package tree as data. It never executes what it reads, and nothing the package says is ever treated as an instruction to the auditor.

    Every finding comes back with a stable key, a severity, a confidence, the exact file span, and a remediation pattern you can hand to the author.

    What it catches

    24 threat IDs across 8 families:

    • Hidden or conflicting instructions
    • Exfiltration paths
    • Over-broad tool permissions
    • Secrets
    • Obfuscation
    • Install-time execution
    • Undocumented network calls
    • Rug-pull risk

    The differentiator is cross-file contradiction. A package can ship a clean entry point and put the payload in a reference file, a changelog, or an example. This tool reads across files, fixes the precedence order, and tells a contradiction apart from a restatement, a paraphrase, an elaboration, or an addition. A single-file scanner cannot do that by construction.

    The number

    99.7% pooled per-assertion pass rate, with-skill (1159/1163), at auditor@1.0 / corpus@1.0.0 / config@cfg-4f9a1c7f.

    That figure is contract-derived, not measured: the with-skill side follows from the frozen contract and the without-skill side is modeled, and no model was executed. Treat it as evidence that the contract is internally consistent, not as a field result.

    What it is not

    • No live runtime monitoring.
    • No network interception.
    • No dependency CVE scanning.
    • No binary reverse engineering.
    • No H2 detection: a trigger cannot be bound to its effect without running the package, so H2 is reported but unscored in v1.0.

    Fine print

    This skill makes no network calls at audit time and executes nothing. external_urls is declared as an explicit empty list.

    The bundle contains descriptions of the patterns it detects, including the taxonomy's own example signals and the auditor-address definition in scope.md. These are the tool's vocabulary, stated in the frozen contract. self-scan.md records the target set, the declared-specification policy, and the two hits by file.

    Four known defects ship in v1.0, each with a backlog item: one package is not fully deterministic on the false-positive guard (BL-01); one package reported a category clean where it could not be checked (BL-02); E3 is claimed by the taxonomy and unexercised by the corpus (BL-03); and one internal expectation describes two incompatible states for H2 (BL-04). Known limits L-1 through L-6 are listed in full in the README.

    H2 is reported but unscored. Static audit cannot bind a trigger to its effect without running the package, so H2 appears in the report at zero deduction. A package whose only defect is an H2 trigger scores 100.

    The benchmark number is contract-derived, not measured. The 99.7% figure follows from the frozen contract; no model was executed. It is evidence that the contract is internally consistent, not a field result. A measured run is on the roadmap.

    E3 is claimed by the taxonomy and unexercised by the test corpus. Detection is specified; no test package currently proves it fires.

    S3 submissions with a complete baseline are untested. S1 and S2 are exercised with baselines; S3 is exercised only without one.

    Baseline contradiction is specified but not tested. No test package supplies a baseline that the package's own declarations contradict.

    One package is not fully deterministic. On the false-positive guard test, one run in three misclassified a benign addition as a contradiction. The modal outcome is stable; the variance is recorded.

    Four defects ship in v1.0, each with a backlog item: the determinism variance above, one package reporting a category clean where it could not be checked, E3 unexercised, and one internal expectation describing two incompatible states for H2.

    How to use

    1. 1

      Connect your agent to the Agensi MCP

      Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup

    2. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before any payment.

    Reviews

    No reviews yet

    Be one of the first to try it. Every listed skill goes through the trust checks below.

    Security scanned

    Checked by our 8-point scan before listing

    1 use

    Times used to date

    Free forever

    No account required to browse

    Trust & safety

    Security scan

    Security scan passed 1 month ago

    • Free with an Agensi account

    Permissions

    No special permissions declared or detected

    Listed1 month ago

    What's inside

    Frequently Asked Questions

    Supply Chain Guard performs a static, pre-install audit of AI skills, plugins, and MCP configurations to detect 24 unique threat types—such as hidden instructions and over-broad permissions—before you ever execute them.

    Unlike single-file scanners, this tool uses cross-file contradiction detection to identify payloads hidden in reference files, changelogs, or examples that might otherwise appear safe at the entry point.

    No, the tool operates entirely offline with no network calls and no code execution, ensuring that the audit process itself cannot be compromised by the package being scanned.

    You receive a comprehensive audit report for each scanned package, featuring a unified trust score, a final verdict, and specific remediation patterns to address any identified security findings.

    This skill is a static analysis tool for AI components; it does not provide live runtime monitoring, network interception, or binary reverse engineering of compiled files.

    The skill is designed to audit standard AI agent packages, plugin manifests, and Model Context Protocol (MCP) server configurations, making it compatible with most modern agentic ecosystems.

    New in Security