The Agent Skills Directory

[COMMAND_EXECUTION]: The skill invokes multiple local Python scripts to perform data collection.
Evidence includes calls to uv run python .claude/skills/my-context-sync/scripts/gmail_fetch.py, calendar_fetch.py, and github_fetch.py.
The GitHub script is executed using a hardcoded absolute Windows path: C:/Users/ash/.claude/skills/my-context-sync/scripts/github_fetch.py.
[PROMPT_INJECTION]: The skill is highly susceptible to indirect prompt injection due to its core function of aggregating untrusted external data.
Ingestion points: Reads content from Slack messages, Gmail body text, GitHub issues/PRs, and Notion pages.
Boundary markers: The skill description does not specify the use of delimiters (e.g., XML tags or triple quotes) to separate untrusted content from agent instructions during the "Result Integration" phase.
Capability inventory: The agent has the capability to write files to the local disk and execute sub-tasks based on the aggregated information.
Sanitization: There is no mention of filtering, escaping, or sanitizing the data collected from external sources before it is processed by the LLM.
[CREDENTIALS_UNSAFE]: The skill documentation describes unsafe methods for handling secrets and authentication tokens.
The skill explicitly mentions passing GITHUB_PERSONAL_ACCESS_TOKEN directly via a command-line environment variable prefix, which can expose the token in process lists or shell history.
It notes that google_token.json is stored locally within the skill directory for Gmail and Calendar access.

my-context-sync