Secret-bearing files (.env, keys, credentials) are indexed and returned to agents #2793

Problem

Tracked secret-bearing files (.env, .env.local, *.pem, *.key, id_rsa*, .npmrc, credentials.json, …) are indexed like ordinary source. Their contents then reach the graph and are returned to AI agents by search_code and get_code_snippet, so indexing a repository can leak its credentials into agent context, transcripts and logs.

Repro: commit a .env containing DB_PASS=example_secret_123, index the repo, then search_code for example_secret — the value comes back verbatim.

Proposal

Skip secret-bearing files during discovery by default (in every index mode), keep templates such as .env.example / .sample / .template / .dist indexed, and add an explicit opt-in config key (index_secret_files, default false) for users who want .env in the graph (e.g. compose ${VAR} resolution or the DotEnv grammar).

I'm aware #1079 asks for richer .env support; the opt-in keeps that path open while making the default safe. If maintainers prefer the opposite default (opt-out), I'm happy to flip it.

I have a tested implementation ready (DCO signed-off) and will open a PR referencing this issue.