The deterministic, verifiable grounding layer for AI code work — a zero-dependency signature-and-evidence map that grounds Claude, Cursor, Copilot, Aider, Windsurf, local LLMs & MCP agents against your real code (repo + installed libraries) so they stop h
Allowed — no malicious behavior detected
19 low-signal patterns reviewed and cleared.
AI security review
scanned · by lpm-firewall-ai
The manifest exposes CLI entrypoints and a prepublishOnly check, not an install lifecycle hook. The required integration excerpts support user-directed guidance and configuration capabilities but do not establish a malicious attack chain.
Static reason
One or more suspicious static signals were detected.; previous stored version diff introduced dangerous source
Trigger
A user invokes a package command that generates context, installs MCP configuration, or installs skills.
Impact
No malicious impact was established by the inspected source.
Mechanism
The inspected paths show generated guidance and project-scoped destinations; they do not establish covert credential forwarding or unconsented lifecycle mutation.
Decision evidence
public snapshot
Evidence for AI clean decision
The package exposes command-line tools and has no preinstall, install, or postinstall hook; its prepublishOnly script runs checks.
The generated Claude guidance tells users to add the listed permissions to settings to activate them.
Claude adapter output also presents the permissions as guidance, rather than writing the settings file.
The Codex adapter targets the project AGENTS.md file and uses a generated-content marker.
The tune adapter markers identify existing project artifacts for adapter selection.
MCP installation targets selected client configuration paths, including project settings.
Skill installation targets project-level client files such as Claude skill files.
The project-map analyzer loads a local analyzer module and handles load errors.
Evidence against
The package supports writing agent-related guidance, MCP configuration, and skill files when users invoke relevant commands.
Automated scan signals; the review checked them and cleared this version. Where they are, most severe file first. Code opens a file's flagged lines, with their evidence, below.
Manifest-reachable source captures an API credential, sends it to a fixed unofficial gateway, and persists that redirection in agent or shell configuration.
gen-context.js
'Add the `permissions.allow` array above to `.claude/settings.json` to activate.',