The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface as it ingests and processes untrusted meeting transcripts from external providers.\n
Ingestion points: Meeting transcript data is retrieved via client.get_transcript() and client.get_summary() in SKILL.md (Step 3).\n
Boundary markers: The skill does not define explicit delimiters or 'ignore embedded instructions' warnings for the transcript text before processing.\n
Capability inventory: The skill has the capability to write markdown files to docs/discussions/, update monthly YAML records in .claude/data/discussions/, and initiate the /secondbrain-adr skill workflow.\n
Sanitization: No sanitization or content validation is performed on the transcript text before the agent uses it to identify architectural decisions or generate discussion summaries.\n- [COMMAND_EXECUTION]: The skill's instructions include Python snippets that modify the system path using sys.path.insert(0, '.claude/lib') to load the FirefliesClient from a local directory within the user's project structure.\n- [DATA_EXFILTRATION]: The skill communicates with the official Fireflies.ai API domain (api.fireflies.ai) to fetch meeting data, using an API key retrieved from local environment variables or configuration files.

secondbrain-transcribe