The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external sources that can be controlled by third parties.
Ingestion Points: PR bodies are fetched via gh pr list and GitHub Project item metadata is retrieved via gh project item-list.
Boundary Markers: No explicit delimiters or instructions are used to ignore potentially malicious embedded content in PR descriptions or issue titles.
Capability Inventory: The skill has extensive shell access (Bash), the ability to execute repository scripts (init.sh), and package management capabilities (pip, pnpm).
Sanitization: While jq and grep -oE are used to filter output, the extracted data (like issue numbers) is interpolated directly into shell loops and commands.
[DYNAMIC_EXECUTION]: The skill executes local source code found within the repository during the bootstrap phase.
Evidence: It explicitly checks for and executes ./scripts/init.sh if present.
Context: While common in development workflows, this pattern allows a repository to define arbitrary commands that the agent will execute during session startup.
[PERSISTENCE]: The skill implements a mechanism to maintain autonomous state across different execution sessions.
Evidence: It checks a memory entity named ActiveOrchestration to determine if a previous task was interrupted.
Behavior: If detected, the skill is instructed to "Resume orchestration immediately
do not wait for user input," which bypasses standard user confirmation for new session activities.
[COMMAND_EXECUTION]: The skill makes heavy use of system CLI tools to manage the environment and project state.
Tools: Uses git, gh (GitHub CLI), docker-compose, and jq to perform operations like starting services, managing branches, and querying project boards.
[EXTERNAL_DOWNLOADS]: The skill automates the installation of dependencies from official registries.
Evidence: It runs pnpm install and pip install during the environment bootstrap step (Step 6).
Trust Scope: These operations target standard package registries (npm and PyPI) for project setup purposes.

session-start