·
Last changedVersion:
This version did not pass the hardening gate. Hardening updated the skill content, deleted the lines high-severity poisoning findings flagged in 4 files before any model saw the skill, and independently checked the resulting changes. Classified findings before hardening: 56. Classified findings closed: 4. Classified findings remaining: 89 (37 introduced by hardening). Unclassified scanner findings remaining: 1. Review findings closed: 22. Review findings remaining: 2.
9
10
37
Expand | File | Group | Subtype | Severity | Status |
|---|---|---|---|---|---|
shared/model-migration.md | Prompt Injection | Direct prompt injection | High | Closed | |
shared/anthropic-cli.md | Excessive Agency & Rogue Autonomy | — | High | Introduced | |
shared/managed-agents-webhooks.md | Sensitive Data & Secret Exposure | Data exfiltration | High | Open | |
shared/managed-agents-tools.md | Privilege Escalation & Host / Cloud Compromise | Privilege escalation | High | Introduced | |
shared/model-migration.md | Sensitive Data & Secret Exposure | Data exfiltration | High | Introduced | |
python/claude-api/tool-use.md | Safety Bypass & Harmful Content | — | High | Introduced | |
typescript/claude-api/README.md | Sensitive Data & Secret Exposure | Data exfiltration | High | Introduced | |
typescript/claude-api/tool-use.md | Safety Bypass & Harmful Content | — | High | Introduced |
93 findings
No skill file found.
Bring Chainguard's hardening pipeline to your first-party skills — custom rulesets, automated remediation, full audit trail.
skills.cgr.dev/public/github.com/anthropics/skills/claude-api:latest
Apache-2.0
HARDENING.md