The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes Node.js scripts to execute git and oh-gc commands, which is necessary for managing repositories and pull requests.
Evidence: Scripts like scripts/create_isolated_issue_commit.mjs, scripts/preflight_commit_scope.mjs, and scripts/scan_existing_submit_targets.mjs use child_process.execFileSync to perform Git operations.
Mitigation: The scripts incorporate path validation logic; for instance, scripts/create_isolated_issue_commit.mjs uses assertRelative to prevent directory traversal and isForbidden to filter sensitive paths such as .ace-outputs/ or binary files.
[DATA_EXFILTRATION]: The skill performs outbound network operations to GitCode for pushing branches and creating pull requests.
Evidence: The workflow involves pushing to personal fork branches and interacting with the GitCode API via the oh-gc CLI.
Safety Controls: The skill enforces a strict "Public Text Redaction" policy defined in SKILL.md (Rules 5 and 6). The scripts/validate_public_text.mjs script is used to scan all generated content for tokens, absolute local paths, and internal directory patterns before they are committed or submitted to the remote server.
[PROMPT_INJECTION]: The skill is subject to an indirect prompt injection attack surface because it processes and summarizes external patch and vulnerability data.
Ingestion points: The skill reads Chromium issue details and patch analysis results from local JSON files (06_merge_result.json, 10_build_fix.json) as described in SKILL.md and processed by scripts/compute_submit_scope.mjs.
Boundary markers: The skill uses explicit templates (e.g., references/pr-body-template.md, references/commit-message-template.md) to structure its output and instructs the agent to stick to a predefined "submit scope."
Capability inventory: The agent has the capability to write to the local Git repository and perform network operations to create Issues and PRs on GitCode.
Sanitization: The scripts/validate_public_text.mjs script provides a layer of protection by redacting sensitive system information, though it does not explicitly filter for malicious instructional patterns in the ingested data.

arkweb-security-patch-gitcode-pr