The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute shell commands to determine environment parameters and format metadata.
Evidence: Uses basename $(pwd) to automatically derive project identifiers ('slugs') from the current working directory.
Evidence: Executes python3 -c "from datetime import datetime; print(datetime.now().astimezone().isoformat(timespec='seconds'))" to generate formatted timestamps for file metadata.
[DATA_EXFILTRATION]: The skill persists potentially sensitive information to the local file system in an insecure manner.
Evidence: Automatically extracts and writes "key conclusions," "rejected directions," and "user complaints" to plaintext files under the ~/.dbs/sessions/ directory.
Risk: The skill instructions explicitly acknowledge that the storage is unencrypted, creating a data exposure risk for commercial secrets or personal information handled during the diagnosis.
[INDIRECT_PROMPT_INJECTION]: The skill creates a persistence mechanism for untrusted user input that could influence the agent in future sessions.
Ingestion points: Conversation history, specifically raw user grievances and diagnostic dialogue.
Boundary markers: The skill uses Markdown headers (e.g., ## 用户主诉) and YAML frontmatter as structure, but lacks instructions for future agents to ignore or escape embedded instructions within those fields.
Capability inventory: The skill utilizes file system writes (mkdir -p, file creation) and shell execution.
Sanitization: No sanitization, escaping, or filtering of the user's original words is performed before persisting the content to disk.

dbs-save