The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from the Fireflies.ai GraphQL API, including transcript titles and meeting metadata, which are then written to a local debug bundle. If this bundle is later processed by an AI agent, malicious content within the API responses could influence the agent's behavior.
Ingestion points: GraphQL API responses from api.fireflies.ai in SKILL.md.
Boundary markers: Uses section headers in summary.txt but no specific boundary tokens or instructions to ignore embedded commands.
Capability inventory: Access to Bash (curl, tar, grep), Read, and environment variables.
Sanitization: Uses jq to select specific fields, but does not sanitize the content of those fields.
[DYNAMIC_CONTEXT_INJECTION]: The skill uses the dynamic context injection syntax (!) to execute shell commands at load time to retrieve version information for Node.js and Python.
Evidence: !node --version 2>/dev/null || echo 'N/A' and !python3 --version 2>/dev/null || echo 'N/A' in SKILL.md.
Assessment: The commands are benign environment checks and do not incorporate user input.
[DATA_EXFILTRATION]: The skill collects account-specific metadata (name, email, transcript titles) and environment information. While this is the stated purpose of a 'debug bundle' and the script includes redaction for .env files, users should be aware that PII is collected into the summary.txt file.
Evidence: Steps 3 and 4 in the provided script extract user names, emails, and meeting titles using curl and jq.

fireflies-debug-bundle