trustcard: Skill for Claude Code

Testing the trustcard behavioral verification engine

Use this skill when you need to verify the lib/behavior.js engine or the mcp-trustcard behavior CLI subcommands on the devin/behavior-engine branch.

Devin Secrets Needed

None. All commands run locally with Node >=18 and the in-repo fixture server.

One-time environment

• Repo root: /home/ubuntu/repos/trustcard (or the current repo root).
• Branch: devin/behavior-engine.
• Node >=18, npm installed, no external dependencies required (package.json has none).
• npm test must pass before claiming behavioral tests are valid.

Commands

Run the full suite:

npm test

Run the behavioral CLI against the benign fixture (process-substitution manifest):

node bin/mcp-trustcard.js behavior \
  <(echo '{"server":{"cmd":"node","args":["test/fixtures/behavior-mem-server.js"]}}') \
  --corpus /tmp/tc-corpus/benign --json > /tmp/tc-benign.json
# Expected: exit 0, report.summary === "pass"

Run the behavioral CLI against the malicious same-contract fixture:

node bin/mcp-trustcard.js behavior \
  <(echo '{"server":{"cmd":"node","args":["test/fixtures/behavior-mem-server.js"],"env":{"TRUSTCARD_EVIL":"1"}}}') \
  --corpus /tmp/tc-corpus/evil --json > /tmp/tc-evil.json
# Expected: exit 1, report.summary === "fail",
# findings include prompt_injection, exfiltration_instruction,
# unexpected_tool_behavior, unexpected_network_attempt, all high severity.

Diff the two reports:

node bin/mcp-trustcard.js behavior diff \
  /tmp/tc-benign.json /tmp/tc-evil.json --json > /tmp/tc-diff.json
# Expected: exit 1, toolsetDigestMatch === true,
# added findings contain the four classes above.

Isolate a single probe (use --probe <tool>:<type>:<index>):

node bin/mcp-trustcard.js behavior \
  <(echo '{"server":{"cmd":"node","args":["test/fixtures/behavior-mem-server.js"]}}') \
  --probe greet:valid:0 --json
# Expected: exit 0, probesTotal === 1, probesPassed === 1, summary === "pass".