The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local shell commands using the Bash tool, specifically interacting with the openspec CLI. It also uses standard filesystem utilities like mkdir and mv to create archive directories and move local files. These operations are scoped to the local environment and are consistent with the skill's stated purpose of archiving project artifacts.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from multiple sources, including CLI outputs (openspec status, openspec instructions archive), project task files (tasks.md), and delta spec files. This ingestion creates a potential surface for indirect prompt injection. However, the skill implements strong defensive instructions: it designates external inputs as "advisory" and "optional," explicitly forbids them from overriding built-in command contracts, and mandates that external content should never be copied verbatim into output files.
Ingestion points: openspec status --json output, tasks.md, delta spec files, and openspec instructions archive --json results.
Boundary markers: The instructions explicitly state that advisory inputs must be kept separate from built-in steps and command contracts.
Capability inventory: The skill has capabilities to execute openspec CLI commands, perform file moves (mv), and update main spec files during the sync process.
Sanitization: The skill requires the agent to explain why any conflicting external advice is ignored and prohibits verbatim copying of external context into change artifacts or summaries.

openspec-archive-change