The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The script scripts/enumerate-uploader.py is designed to read sensitive authentication tokens from ~/.bilibili-cli/credential.json. Accessing private credentials from the configuration directories of other CLI tools is a high-risk data exposure pattern.
[INDIRECT_PROMPT_INJECTION]: The transcription cleanup scripts (scripts/clean-transcripts.py, scripts/dual-transcribe.py, scripts/reclean-en.py) ingest untrusted video titles and ASR transcripts and interpolate them into LLM prompts. This allows malicious video content or metadata to potentially influence the agent's logic during the cleanup phase.
Ingestion points: Ingests Bilibili video titles from metadata.json and generated ASR text from transcript.md.
Boundary markers: Absent. The prompt templates do not use safety delimiters or instructions to ignore embedded commands.
Capability inventory: The skill possesses capabilities for subprocess execution (scripts/transcribe.sh, scripts/pipeline-queue.py) and file system access.
Sanitization: While titles are slugified for filesystem safety, no sanitization is applied for content interpolated into LLM prompts.
[COMMAND_EXECUTION]: Multiple scripts (scripts/transcribe.sh, scripts/pipeline-queue.py, scripts/collect-metrics.py) make extensive use of subprocess.run to execute shell commands and external utilities. These calls incorporate external data such as video IDs and directory paths, increasing the risk of command injection if inputs are not strictly validated.
[DYNAMIC_EXECUTION]: The test suite (tests/test_chunk_transcribe.py, tests/test_dual_transcribe.py) utilizes importlib.util to dynamically load and execute script files as modules from computed paths.

bilibili-transcriber