SKILLFORGE
Agent context security

SCAN THE FILESYOUR AGENTSSILENTLY TRUST

SkillForge audits SKILL.md, AGENTS.md, CLAUDE.md, Cursor rules, Copilot instructions, and MCP configs before your team lets Claude, Codex, Cursor, or agent harnesses read them.

ClaudeCodexCursorMCPPaid scanner
View Sample ReportStart Pro - $29/mo
Need to create a skill? The builder is still available.

Full scans are paid. The sample report shows the exact inventory, findings, verdict, and mitigation format before checkout.

agent-context-report.md
review
Agent Context Risk Report
Repository: acme/agent-workflows
Verdict: NEEDS TEAM REVIEW
 
Inventoried files
- AGENTS.md                  Codex
- CLAUDE.md                  Claude
- .claude/skills/deploy/SKILL.md
- .cursor/rules/release.mdc  Cursor
- .mcp.json                  MCP
 
Findings
[HIGH] Outbound curl in deploy skill
       Could exfiltrate release notes
       or environment values.
 
[MEDIUM] Broad MCP filesystem access
         Add path allowlists before use.
 
Next step: manual review before sharing
these instructions across the team.
5 files inventoried2 findings

How it works

FROM REPO URL TO
TRUST DECISION

01

Inventory

Paste a public GitHub repo or upload a ZIP. SkillForge discovers the agent files that Claude, Codex, Cursor, Copilot, and MCP can trust.

AGENTS.md, CLAUDE.md, SKILL.md, .mcp.json

02

Analyze

Static rules and AI review check for hidden instructions, shell execution, exfiltration, broad permissions, secrets, and unsafe installs.

safe / review / block verdict

03

Remediate

Download a report, fix risky instructions, and rescan through Pro or Team before approving the files for team use.

Markdown report for PRs and security review

Why teams pay

AGENT FILES ARE
SUPPLY CHAIN NOW

Your agents can read instructions, run tools, install packages, and touch repositories. SkillForge gives teams a review layer before those instructions become shared behavior.

Repo-wide context discovery

Finds SKILL.md, AGENTS.md, CLAUDE.md, Cursor rules, Copilot instructions, and MCP configs in one scan.

Agent-specific risk checks

Flags prompt injection, command execution, exfiltration, secrets, obfuscation, broad MCP permissions, and unsafe installs.

Platform influence map

Shows which files affect Claude, Codex, Cursor, Copilot, MCP, or generic agent harness behavior.

Clear install verdict

Every report ends with safe to trust, needs team review, or block until fixed.

Shareable risk report

Download a Markdown report with file inventory, severity findings, and remediation guidance.

Team review workflow

Keep scan history, share reports, and apply consistent policy thresholds across team repositories.

Agent Context Scanner

SEE WHAT YOUR AGENTS
WILL TRUST BY DEFAULT

Coding agents now load repo instructions, skills, rules, and MCP configs as trusted context. SkillForge inventories those files and flags behavior your team should review before adoption.

Agent Context Risk Report
7.1

/10.0

Needs Team Review

5 files inventoried - 3 findings detected

highOutbound curl in .claude/skills/deploy/SKILL.md
mediumMCP filesystem server has no path allowlist
infoAGENTS.md affects Codex in nested workspaces

Why this matters

These files can change how an agent uses tools, reads files, handles secrets, and executes shell commands. Review them like dependencies before sharing across a team.

Repo-level discovery

Scan the source, not just one uploaded file. The report maps every trusted agent instruction file it can find.

Static checks + AI review

Deterministic rules catch obvious dangerous patterns; AI review explains subtle risks in natural language.

Paid team workflow

Pro and Team plans unlock full scans, exports, scan history, and mitigation guidance for repeated review.

What we check

Prompt InjectionCommand ExecutionNetwork ExfiltrationSecret AccessFile System MutationExcessive PermissionsObfuscated PayloadsUnsafe InstallsHidden Persistence
View Sample ReportSee pricing

Pricing

PAID FROM DAY ONE,
BECAUSE TRUST IS THE PRODUCT

View the sample report before subscribing. Use Pro or Team when you need real repo scans, exports, and mitigation guidance.

Start here

Pro

For developers and small teams who want the full report.

$29per month
  • Repo and ZIP agent context scans
  • Full Agent Context Risk Report
  • Markdown export
  • Mitigation recommendations
  • Early access to API experiments

Team

For teams standardizing Claude, Codex, Cursor, or MCP.

$99per month
  • Up to 5 active repos
  • Scan history
  • Shareable reports
  • Team policy thresholds
  • Priority scanner improvements

Enterprise

For private repo workflows and procurement-heavy buyers.

Customsecurity review
  • Private repo review
  • Custom detectors and policies
  • Audit export support
  • Security questionnaire support
  • SSO and admin controls roadmap
Send enquiry

FAQ

Frequently Asked Questions

Product

Scanner

BEFORE YOUR TEAM
TRUSTS THE AGENT

Scan the instructions, rules, skills, and MCP configs that shape agent behavior. Then decide: safe, review, or block.

View Sample ReportStart Pro

Works with Claude, Codex, Cursor, Copilot instructions, and MCP configs.