diff --git a/.gitleaks.toml b/.gitleaks.toml new file mode 100644 index 0000000..cd3ab63 --- /dev/null +++ b/.gitleaks.toml @@ -0,0 +1,43 @@ +# SPDX-License-Identifier: MPL-2.0 +# +# Gitleaks configuration — hyperpolymath/echidna +# +# EXTENDS THE ESTATE BASELINE, it does not replace it. +# `.gitleaks-estate.toml` is staged into the workspace root by +# hyperpolymath/standards `.github/workflows/secret-scanner-reusable.yml` +# before the scan runs. gitleaks resolves `[extend] path` against the process +# CWD (verified — NOT relative to this file), which is the repository root. +# +# WHY THIS FILE IS LOCAL RATHER THAN IN THE ESTATE BASELINE. +# Every entry below is a blind spot. Held here, it blinds this repository only, +# and the justification sits beside the code it describes. Promoted to the +# estate baseline it would blind all 400+ repositories — so the baseline keeps +# only entries that are true everywhere (lockfiles, vendored bundles, published +# protocol constants). +# +# Each entry names WHAT THE VALUE IS. "This file is noisy" is not a reason; if +# an entry cannot say what the matched value actually is, the secret should be +# removed from the tree instead. +# +# Every finding suppressed here was triaged on 2026-08-06 by reading the +# matched line with the value redacted. Before adding an entry, plant a +# realistic secret in the same path and confirm it is STILL detected. + +[extend] +path = ".gitleaks-estate.toml" + +[allowlist] +description = "hyperpolymath/echidna: locally justified exemptions, extending the estate baseline" + + +paths = [ + # Declared machine-learning corpus. These JSONL files are training data + # FOR a security/proof analyser: containing credential-shaped strings is + # their entire purpose. 69 of this repo's 73 findings are this directory. + '''(^|/)training_data/[^/]*\.jsonl$''', + + # Model vocabulary file — a token list extracted from the corpus above, so + # it inherits the same credential-shaped strings by construction. + '''(^|/)models/premise_vocab\.txt$''', + +]