GitHub - VihaanInnovations/autonoma: Detects and auto-fixes hardcoded secrets in Python repos — refuses when the fix could break your code

Autonoma detects hardcoded secrets and replaces them with os.environ[...] references using AST rewrites. Changes are applied only when the rewrite is deterministic and semantically-preserving.

Rewrites use the parsed syntax tree — not regex on raw text. Runs locally with no network calls. Works in CI with idempotent, minimal diffs.

What problem this solves

Hardcoded secrets in codebases:

• secrets get committed and stay in git history
• fixing them manually breaks code or misses edge cases
• teams detect leaks but avoid auto-fix tools because they are unsafe

Most tools detect them.
Autonoma only touches what it can prove is safe to change.

Quick example

autonoma scan .
autonoma fix .
git diff

Installation

Pre-commit Integration

Add this to your .pre-commit-config.yaml to prevent secrets from entering your history:

- repo: local
  hooks:
    - id: autonoma
      name: Autonoma Scan
      entry: autonoma scan
      language: system
      types: [python]

Try it in 60 seconds

# 1. Create a test file with hardcoded secrets
cat > test_secrets.py << 'EOF'
SENDGRID_API_KEY = "sg-live-abc123xyz789"
DB_PASSWORD = "Pr0dAccess2024!"
EOF

# Also create the env contract file (required for safe remediation)
printf 'SENDGRID_API_KEY=\nDB_PASSWORD=\n' > .env.example

# 2. Scan — emits JSON findings to stdout
autonoma scan test_secrets.py

# 3. Fix — rewrites the file in place
autonoma fix test_secrets.py

# 4. Scan again — should now be clean (exit 0)
autonoma scan test_secrets.py

# 5. Inspect the result
cat test_secrets.py

Expected output after fix:

import os
SENDGRID_API_KEY = os.environ["SENDGRID_API_KEY"]
DB_PASSWORD = os.environ["DB_PASSWORD"]

Commands

scan

Detection mode. Outputs JSON to stdout and a human-readable summary to stderr. Non-mutating — never modifies files.

# Scan a directory (JSON findings to stdout)
autonoma scan src/

# Save JSON results to a file
autonoma scan src/ > findings.json

Exit codes for scan:

fix

Remediates hardcoded secrets using AST rewrites. Mutates files in place.

# Apply fixes
autonoma fix src/

# Preview patches before writing
autonoma fix src/ --diff

# Write remediation audit log
autonoma fix src/ --report-out audit.json

Exit codes for fix:

The fix command exits 1 whenever it found secrets before attempting remediation, regardless of whether the rewrite succeeded. This is intentional: CI pipelines should flag the commit where secrets were introduced, even after auto-fix. Run autonoma scan afterward to confirm the repo is clean.

history-scan

Scans git history for secrets that were committed and later removed or changed.

Before / After

These are the patterns Autonoma actually fixes today.

Before

# settings.py
SENDGRID_API_KEY = "sg-live-abc123xyz789"
DB_PASSWORD = "Pr0dAccess2024!"

After (autonoma fix .)

# settings.py
import os
SENDGRID_API_KEY = os.environ["SENDGRID_API_KEY"]
DB_PASSWORD = os.environ["DB_PASSWORD"]

Refused (refusal-first safety)

# f-string — refused because the rewrite would change semantics
api_key = f"prefix_{BASE_KEY}"
# → REFUSED: refuse_fstring_mixed_expression

# Dict/nested value — refused because the target is not a simple assignment
DATABASES = {
    "default": {"PASSWORD": "Pr0d@ccess2024!"}
}
# → REFUSED: unsupported assignment target type

Refused findings are reported in the JSON output and cause a non-zero exit in CI. Files with refused findings are never modified.

What Autonoma fixes vs refuses

CI/CD Features

Re-running on a clean file makes no changes. Rewrites keep original indentation and comments. import os is added only if it's missing.

Integration & CI/CD

GitHub Actions (Scan Only)

To fail your build if any secrets are detected:

- name: Scan for secrets
  run: autonoma scan .

Legacy Commands

analyze is retained for backwards compatibility. Migrate to scan or fix.

# Equivalent to 'autonoma scan'
autonoma analyze src/ --detect-only

# Equivalent to 'autonoma fix'
autonoma analyze src/ --auto-fix

Constraints & Behaviors

What it remediates

• Simple assignments: API_KEY = "secret"
• Class attributes: class Config: PASS = "secret"
• Keyword arguments: connect(password="secret")

What it refuses (by design)

• Complex Expressions: f-strings, concatenations, or function calls on the RHS.
• Ambiguous Targets: Multiple assignments (A = B = "secret") or tuple unpacking.
• Nested/Dict Values: Values inside dicts, lists, or tuples.
• Missing Context: If no .env.example or environment contract is found in the repo.

Refused cases are reported in JSON output and will cause non-zero exit codes in CI. The file is never modified if any issue in it is refused.

What it does not do

• It does not use entropy/guessing (it uses heuristic name matching).
• It does not modify non-Python files in the Community Edition.
• It does not delete your code; backups are written as <file>.bak before modification.

JSON Schema

autonoma scan outputs a detect-only report to stdout:

{
  "schema_version": "1.0",
  "tool_name": "autonoma",
  "tool_version": "0.1.5",
  "generated_at": "2026-03-24T12:00:00Z",
  "mode": "detect-only",
  "summary": {
    "files_processed": 3,
    "total_findings": 2,
    "safe_to_fix": 1,
    "refused": 1
  },
  "findings": [
    {
      "file": "settings.py",
      "line": 4,
      "pattern_type": "api_key",
      "severity": "high",
      "rule_id": "SEC002",
      "safe_to_fix": true,
      "suggested_env_var": "SENDGRID_API_KEY",
      "refusal_reason": null,
      "fingerprint": "sha256:abc123..."
    }
  ]
}

License

MIT License