The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local shell scripts and command-line utilities to discover accounts and verify test environments.
Evidence: Executes a local vendor script located at ${CLAUDE_PLUGIN_ROOT}/scripts/find-vault-logins.sh to search for 1Password credentials.
Evidence: Utilizes the op (1Password) CLI to read secret fields and jq for JSON processing.
Evidence: Invokes curl to authenticate with project APIs and query for record existence.
[DATA_EXFILTRATION]: The skill accesses sensitive credentials and performs network operations, though it implements significant guardrails.
Evidence: Reads passwords from 1Password vaults using op read.
Evidence: Sends these credentials to specified <api-url> endpoints via curl to establish sessions for data verification.
Mitigation: The skill enforces "Hard Rules" against leaking secrets, tokens, or pre-authenticated links into the final Linear comments and requires deleting session files after the run.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data sources which could be used for indirect injection attacks.
Ingestion points: Reads git diffs, browser walk logs (final_list), and external API responses from the stage environment.
Boundary markers: The skill does not explicitly mention using delimiters or sanitization for content ingested from the diff or API responses.
Capability inventory: The agent has access to shell execution, network requests, and vault access.
Sanitization: No explicit validation or filtering of external data content is defined.
[DYNAMIC_EXECUTION]: The skill constructs and executes shell commands dynamically based on environment configuration.
Evidence: Dynamically assembles curl commands and authentication payloads using variables like <api-url>, <login>, and vault identifiers found during the account resolution phase.

writing-qa-test-instructions