Configuration Guide

Configuration File

licit uses a YAML file called .licit.yaml in the project root. It is automatically generated with licit init but can be edited manually.

Configuration Resolution

When licit loads the configuration, it follows this priority order:

1. Explicit path: licit --config /path/to/config.yaml <command>
2. File in current directory: .licit.yaml in the cwd
3. Defaults: If no file exists, uses default values

If the file exists but has YAML syntax or validation errors, licit logs a warning and uses the defaults.

Full Example

# .licit.yaml — licit configuration
# All fields are optional. Default values are shown.

provenance:
  enabled: true
  methods:
    - git-infer          # Methods: git-infer, session-log, git-ai
  session_dirs: []       # Directories with agent session logs
  sign: false            # Sign records with HMAC-SHA256
  sign_key_path: null    # Path to signing key
  confidence_threshold: 0.6  # Minimum threshold to classify as AI
  store_path: .licit/provenance.jsonl

changelog:
  enabled: true
  watch_files:           # Agent config files to monitor
    - CLAUDE.md
    - .cursorrules
    - .cursor/rules
    - AGENTS.md
    - .github/copilot-instructions.md
    - .github/agents/*.md
    - .architect/config.yaml
    - architect.yaml
  output_path: .licit/changelog.md

frameworks:
  eu_ai_act: true        # Enable EU AI Act evaluation
  owasp_agentic: true    # Enable OWASP Agentic Top 10 evaluation
  nist_ai_rmf: false     # Future (V1)
  iso_42001: false       # Future (V1)

connectors:
  architect:
    enabled: false
    reports_dir: .architect/reports
    audit_log: null
    config_path: null
  vigil:
    enabled: false
    sarif_path: null
    sbom_path: null

fria:
  output_path: .licit/fria-report.md
  data_path: .licit/fria-data.json
  organization: ""
  system_name: ""
  system_description: ""

annex_iv:
  output_path: .licit/annex-iv.md
  organization: ""
  product_name: ""
  product_version: ""

reports:
  output_dir: .licit/reports
  default_format: markdown   # markdown, json, html
  include_evidence: true
  include_recommendations: true

Detailed Sections

provenance — Code Traceability

Controls how licit tracks code origin (AI vs human).

Available methods:

• git-infer: (Implemented) Analyzes git history heuristics with 6 signals: author pattern, commit message, change volume, co-authors, file patterns, and time.
• session-log: (Implemented) Reads AI agent session logs. Currently supports Claude Code (JSONL files in ~/.claude/projects/). Extensible via the SessionReader Protocol.
• git-ai: Reads annotations from git-ai type tools (planned).

Example — Enable signing:

provenance:
  sign: true
  sign_key_path: ~/.licit/signing-key

Example — Multiple methods:

provenance:
  methods:
    - git-infer
    - session-log
  session_dirs:
    - ~/.claude/projects/

changelog — Agent Config Monitoring

Status: Functional since v0.3.0. Run with licit changelog.

Tracks changes in AI agent configuration files through git history, producing semantic diffs with severity classification (MAJOR/MINOR/PATCH).

Default monitored files:

Patterns with * are resolved using Path.glob(). Exact names check for existence in git history.

Processing pipeline:

ConfigWatcher → Semantic Differ → ChangeClassifier → ChangelogRenderer
  (git log)     (YAML/JSON/MD)   (MAJOR/MINOR/PATCH)   (MD/JSON)

Diff formats: YAML and JSON produce field-level diffs (model, llm.provider). Markdown produces section-based diffs (section:Rules). Plain text produces a full content diff.

Output formats: markdown (default) groups by file and sorts by severity. json produces a {"changes": [...]} object.

Example — Add a custom file:

changelog:
  watch_files:
    - CLAUDE.md
    - .cursorrules
    - my-custom-agent.yaml       # additional file
    - .prompts/**/*.md             # recursive glob

Example — JSON output:

licit changelog --format json --since 2026-01-01

For detailed documentation, see Changelog System.

frameworks — Regulatory Frameworks

Controls which regulatory frameworks are evaluated.

Example — EU AI Act only:

frameworks:
  eu_ai_act: true
  owasp_agentic: false

connectors — External Integrations

Status: Functional since v0.7.0. Connectors are formal: they implement the Connector Protocol, handle errors gracefully, and report results via ConnectorResult.

Connectors are optional — licit works fully without them. When enabled, they enrich the EvidenceBundle with additional data that improves the compliance evaluation.

Auto-detection: licit init automatically detects the presence of architect (.architect/) and vigil (.vigil.yaml) and enables the corresponding connectors.

Inline fallback: When no config (LicitConfig) is available, EvidenceCollector builds temporary connectors with auto-detected paths. This guarantees backwards compatibility.

connectors.architect

Integration with Architect to read audit reports, audit logs, and guardrail configuration.

Data sources it reads:

Config YAML fields extracted:

# .architect/config.yaml
guardrails:
  protected_files: [.env, secrets.yaml]    # → guardrail_count += N
  blocked_commands: [rm -rf]               # → guardrail_count += N
  code_rules: [no-eval]                    # → guardrail_count += N
  quality_gates: [lint, test]              # → has_quality_gates, quality_gate_count
costs:
  budget_usd: 50.0                         # → has_budget_limits
dry_run: true                              # → has_dry_run (default True if absent)
rollback: true                             # → has_rollback (default True if absent)

Full example:

connectors:
  architect:
    enabled: true
    reports_dir: .architect/reports
    config_path: .architect/config.yaml
    audit_log: .architect/audit.jsonl

connectors.vigil

Integration with Vigil and other security scanners that produce SARIF 2.1.0.

SARIF path resolution:

1. Explicit sarif_path: if it’s a file, reads it; if it’s a directory, reads all *.sarif files.
2. Auto-detected: .sarif files found by ProjectDetector in the project.
3. Deduplication: if the same file appears in both sources, it is read only once.

SARIF severity mapping:

Note: All runs in the SARIF are parsed, regardless of the tool name (vigil, Semgrep, CodeQL, etc.).

Full example:

connectors:
  vigil:
    enabled: true
    sarif_path: reports/security/    # Directory with .sarif files
    sbom_path: sbom.json             # CycloneDX SBOM (V1: will feed OWASP ASI03)

fria — Fundamental Rights Impact Assessment

Configuration for the FRIA (EU AI Act Article 27).

annex_iv — Annex IV Technical Documentation

reports — Report Generation

Data Directory (.licit/)

licit stores all its internal data in the .licit/ directory within the project root.

.licit/
├── .signing-key        # HMAC-SHA256 key (auto-generated if sign=true)
├── provenance.jsonl    # Traceability store (JSONL append-only)
├── changelog.md        # Agent config changelog
├── fria-data.json      # Raw FRIA data
├── fria-report.md      # Human-readable FRIA report
├── annex-iv.md         # Annex IV technical documentation
└── reports/            # Generated reports
    ├── provenance.md          # Provenance report
    ├── compliance-report.md
    ├── compliance-report.json
    └── compliance-report.html

Recommended .gitignore:

# licit — internal data (may contain sensitive information)
.licit/provenance.jsonl
.licit/fria-data.json

# licit — generated reports (commit if desired)
# .licit/reports/

It is recommended to commit .licit.yaml and the generated reports, but not the provenance store or raw FRIA data, as they may contain sensitive team information.

Programmatic Usage

The configuration can be loaded and manipulated from Python:

from licit.config.loader import load_config, save_config

# Load
config = load_config()  # Automatically looks for .licit.yaml
config = load_config("/explicit/path/.licit.yaml")

# Modify
config.frameworks.owasp_agentic = False
config.connectors.architect.enabled = True

# Save
save_config(config)  # Saves to .licit.yaml
save_config(config, "/another/path/config.yaml")