The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local Python scripts (credential-manager.py, approval-manager.py, and execution-tracker.py) via the command line. These scripts are used to validate credentials, manage user approvals for high-risk actions, and maintain an audit log of client switches.
[CREDENTIALS_UNSAFE]: The skill accesses sensitive file paths within ~/.claude-marketing/credentials/ and ~/.claude-marketing/brands/. These files contain API keys, environment variables, and OAuth tokens for various marketing platforms. This access is the core functionality of the skill, allowing it to switch contexts and validate access before executing tasks.
[INDIRECT_PROMPT_INJECTION]: The skill ingests data from local brand profile files and user-supplied strings, which are then passed to system scripts, creating a potential attack surface.
Ingestion points: Brand profile data is read from ~/.claude-marketing/brands/{slug}/profile.json, and user input is accepted for brand slugs and switch reasons.
Boundary markers: No specific delimiters or boundary markers are defined to isolate untrusted data from the instructional context.
Capability inventory: The skill has the capability to execute subprocesses (Python scripts) and perform file read/write operations within its configuration directory.
Sanitization: The instructions do not specify any sanitization or validation routines for the data read from the profiles or the input provided by the user before they are processed by internal scripts.

credential-switch