The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes project-specific data to generate prompts ("briefs") for downstream agents, creating a surface for indirect instructions.
Ingestion points: Accesses configuration files such as .conductor/settings.toml, .conductor/settings.local.toml, and conductor.json within user repositories.
Capability inventory: Executes CLI commands (conductor), queries local SQL views, interacts with the api.conductor.build API, and opens system deep links.
Sanitization: Implements percent-encoding for URL construction via Python's urllib.parse.quote, but does not specify content sanitization for the data ingested from repository files.
Boundary markers: Lacks explicit delimiters or instructions for the agent to ignore potentially malicious embedded content within the ingested configuration files.
[COMMAND_EXECUTION]: Utilizes system commands and a specific CLI tool to manage workspaces and sessions.
Executes the open command to trigger conductor:// deep links on macOS.
Interacts with the Conductor CLI, typically located at /Applications/Conductor.app/Contents/Resources/bin/conductor.
Provides a utility for running SQL queries against local transcripts via conductor sql.
[EXTERNAL_DOWNLOADS]: References and interacts with external infrastructure associated with the Conductor.build service.
Communicates with the public API at api.conductor.build.
Directs users to official documentation at conductor.build.
These resources are integral to the stated purpose of the skill and belong to the tool's official vendor infrastructure.
[DYNAMIC_EXECUTION]: Generates and executes a short Python script to perform URL encoding for deep links.
Assembles a Python command string using python3 -c to invoke urllib.parse.quote on workspace briefs and repository paths.
[DATA_EXFILTRATION]: Instructions involve the handling of environment configuration files.
Identifies and processes .env files as part of the workspace setup and "files to copy" mechanism.
Accesses user and project-level settings in ~/.conductor/settings.toml and repo-local .conductor/ directories.
The skill includes a positive security control by explicitly instructing the agent not to collect or ask for authentication tokens directly from the user.

conductor