feat(security): config secret redaction — Spring application.{properties,yml} + Shopify {% schema %} (closes #58 phase-2)

[Wolfvin]

Summary

Implements Phase 2 of issue #58 — config secret redaction for Spring application.properties/application.yml and Shopify Liquid {% schema %} blocks.

(Supersedes the closed PR #133 — that PR was based on the Phase 1 branch feat/issue-58-path-traversal-phase-1, which got deleted when PR #131 was merged. This PR is rebased directly onto main and contains only the Phase 2 commit.)

CodeLens's job is to map code structure — not to be a vault for production secrets. But many config-file formats mix structural metadata (key names) with sensitive values (passwords, API keys, connection strings) in the same file. If an agent-driven extraction layer indexes these files naively, CodeLens becomes a secret sink — the value ends up in the SQLite registry, outline.json, MCP tool responses, and snapshot exports.

This PR provides format-aware redaction that returns a key-only view of these config files. The raw file content stays on disk — an agent that needs the actual value reads the file directly, just like any source file. CodeLens never persists it.

Depends on #131 (Phase 1, MERGED) — Phase 2's safe_read_file_for_indexing composes Phase 1's safe_read_file_within_project with the new redaction layer.

Files

New:

• scripts/security/config_secret_redaction.py — redact_application_properties(), redact_application_yml(), redact_shopify_schema(), redact_config_file() (auto-detect), is_config_file(), safe_read_file_for_indexing() (composes Phase 1 path-traversal + Phase 2 redaction)
• tests/test_config_secret_redaction.py — 63 tests covering all three formats, idempotency, malformed-JSON handling, profile-specific Spring variants, integration with Phase 1 path traversal

Modified:

• scripts/security/__init__.py — re-exports Phase 2 public API alongside Phase 1

Key design decisions

1. Conservative bias — when in doubt, redact. A false positive (redacting a non-secret value) costs an agent a file read; a false negative (storing a real secret) costs a security incident. The structural allowlist (server.port, spring.datasource.url, spring.application.name, etc.) is intentionally narrow.
2. Line-by-line YAML parsing, not PyYAML. PyYAML is an optional dependency (Phase 1 doctor warns when missing); making it required for redaction would break CodeLens on minimal installs. Spring application.yml is a strict subset of YAML where every value is on the same line as its key — the line-by-line approach handles it correctly. A full YAML parser would also happily load anchors, tags, and multi-line strings — each one a redaction-bypass waiting to happen.
3. YAML bare-key allowlist, not dotted-path. In YAML the key is the bare last segment (url:, port:, password:) — not the dotted path. The properties-file allowlist matches dotted paths; the YAML allowlist matches bare keys. Both are conservative.
4. Shopify schema: parse JSON, walk in-place, re-serialize. Walks the parsed JSON tree, replacing every default field whose value is a non-empty string with <redacted>. Number/boolean defaults preserved (they're never secrets). Malformed JSON → block replaced with a {% comment %}schema redacted (malformed JSON){% endcomment %} marker so the agent sees something happened.
5. safe_read_file_for_indexing composes Phase 1 + Phase 2. Single safe entry point for any extraction layer that might ingest config files: path-traversal check first (Phase 1), then format-aware redaction (Phase 2), then return.
6. Idempotent. Re-running redaction on already-redacted content is a no-op (verified by tests) — so a future "redact on read AND on write" pipeline won't double-redact.
7. Never raises on format errors. A parse failure in one config file must not break a workspace-wide scan. Falls back to a placeholder comment + log warning.

What Phase 2 deliberately does NOT do

• It does not detect secrets by pattern (that's secrets_engine.py's job, already shipped).
• It does not refuse to read config files (Phase 1 path-traversal handles file-access safety; this module handles value-persistence safety).
• It does not auto-discover config files — callers pass file paths in. The redaction is a pure function on content.

Acceptance criteria

• Spring application.properties values redacted (except structural allowlist)
• Spring application.yml scalar values redacted (except allowlist + numbers/booleans)
• Shopify {% schema %} string default fields redacted
• Profile-specific Spring variants (application-prod.properties) detected
• Malformed Shopify JSON handled safely (block replaced with comment marker)
• Idempotent (re-redaction is a no-op)
• safe_read_file_for_indexing integrates Phase 1 path-traversal + Phase 2 redaction
• No new dependencies (uses stdlib json + re only)

(Phases 3–7 of #58 — git safety guard, Secretlint, file system MCP tools, LLM schema validation, PoC generation — are out of scope for this Phase 2 PR.)

Test results

• 63 new tests in tests/test_config_secret_redaction.py — all pass
• Phase 1 tests (tests/test_path_traversal.py, 41 tests) — still pass (no regression after rebase onto merged main)

Closes #58 (phase-2).

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.

[sonarqubecloud]

Quality Gate failed

Failed conditions
1 Security Hotspot
C Security Rating on New Code (required ≥ A)

See analysis details on SonarQube Cloud

Catch issues before they fail your Quality Gate with our IDE extension SonarQube for IDE