The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill implements a handover mechanism where it scans the local workspace for custom build definitions such as proposal-build scripts, Makefile, or justfile. If detected, the skill halts its default processing and instructs the agent to run the discovered code.
Ingestion points: The scripts/publish.py script performs a directory traversal of the proposal's parent folder to identify executable files or build recipes matching the proposal-build naming convention.
Boundary markers: Absent. The skill does not define specific delimiters or validation logic for the content of the workspace-supplied scripts.
Capability inventory: The skill utilizes subprocess.run to execute pandoc and typst. Additionally, the markdown instructions in SKILL.md grant the agent permission to use shell capabilities to execute untrusted code found in the workspace.
Sanitization: None. The skill relies on the agent to execute code paths discovered in the user's local environment without integrity verification.
[COMMAND_EXECUTION]: The instructions in SKILL.md explicitly mandate that the agent execute arbitrary scripts or targets found in the workspace when a handover condition is met.
Evidence: "When one is found, publish builds nothing and exits 3, naming what it found. Run that definition and relay its output."
Evidence: "Run it with: PROPOSAL_PATH=... [runner] [target]"
[EXTERNAL_DOWNLOADS]: The skill provides installation guidance for external document processing tools. These references point to well-known and reputable software repositories and package managers.
Evidence: Guidance to install pandoc and typst via winget (Microsoft), brew (Homebrew), or official Linux distribution packages.

proposal-publish