The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external sources and instructs the agent to follow directions contained within that data. Specifically, it fetches JSON output from openspec instructions archive and openspec instructions specs to guide the archiving and syncing processes.
Ingestion points: SKILL.md Step 1 and Step 4 use the output of openspec CLI commands to populate context, operationGuidance, and rules fields.
Boundary markers: The skill specifies that the agent should treat context and operationGuidance as prompt-level inputs and provides explicit conflict resolution rules (e.g., "If context conflicts with one of those controlling inputs, report the conflict and preserve the controlling value").
Capability inventory: The skill possesses the ability to execute shell commands (Bash), move directories (mv), and modify project specification files via the openspec-sync-specs workflow.
Sanitization: There is no automated sanitization; the skill relies on instructional guardrails to prevent the agent from blindly following instructions that conflict with command contracts or user choices.
[COMMAND_EXECUTION]: The skill executes shell commands (mkdir, mv) using variables such as <changeRoot> and <target-name> derived from project metadata.
Evidence: Step 5 performs file system modifications using paths retrieved from openspec status --json. While these paths are quoted to prevent basic word splitting, they represent a surface where maliciously crafted directory structures or change names could attempt shell redirection or command injection.

openspec-archive-change