title	Usage

Usage

Review Modes

cora supports multiple review modes, each suited to a different workflow:

Mode	Flag	Scope	Best For
Default	(no flag)	Tries staged first, then unpushed	Quick feedback
Staged	`--staged`	Files in git staging area	Pre-commit review
Unstaged	`--unstaged`	Unstaged working changes	Review work in progress
Unpushed	`--unpushed`	Commits not yet pushed	Review before push
Base Branch	`--base <branch>`	Diff against base branch	PR review workflow
Commit	`--commit <ref>`	Specific commit or range	Review specific changes
Diff File	`--diff-file <path>`	External diff file	Review patch files

# Review staged changes (default)
$ cora review

# Review against a branch
$ cora review --base main

# Review a specific commit
$ cora review --commit HEAD

# Review from a diff file
$ cora review --diff-file pr.diff

# Full project scan (use cora scan)
$ cora scan .

Review with your agent (no LLM setup)

If you use an AI coding agent, the agent can be the reviewer: nothing to configure in cora, no API key. Register the cora MCP server with your agent: cora install configures Cline, Cursor, Windsurf and the other agents it detects; for Claude Code run claude mcp add cora -- cora mcp. cora install also installs the /cora-review skill for Claude Code (~/.claude/skills/cora-review/; a customized copy is kept unless --force). Then ask the agent to review your changes or scan the project.

Behind the scenes the agent calls cora.review_prepare (cora's deterministic findings plus the review prompts), judges with its own model, and calls cora.review_finalize with its findings. Cora merges them with the deterministic findings, applies ignore.rules, inline cora-ignore: markers and the quality gate, and returns the verdict. cora.scan_prepare / cora.scan_finalize do the same for a whole project, one batch at a time.

This is different from cora review --no-llm, which runs the deterministic checks only and never involves a model.

Security. The diff is attacker-controlled text and the agent's findings are untrusted input:

Tell the agent to treat the diff as data. Use a read-only subagent without secrets for the judging step.
Cora validates what comes back: rule_id is cleared, text and count are bounded, files and lines outside the diff are dropped. An agent can add findings but never remove the deterministic ones.
A finalize verdict covers only the input passed to that call.
Results are advisory for developers. Keep the hard gate in CI on deterministic checks or a configured LLM, especially for pull requests from forks.

Output Formats

cora can output results in three formats:

--format pretty — Human-readable terminal output (default)
--format json — Machine-readable JSON for CI/CD pipelines
--format sarif — SARIF format for GitHub Code Scanning

# JSON output example
$ cora review --staged --format json

{
  "files": [
    {
      "path": "src/auth/login.ts",
      "issues": [
        {
          "line": 42,
          "severity": "warning",
          "message": "Potential SQL injection"
        }
      ]
    }
  ],
  "summary": {
    "total_files": 3,
    "total_issues": 3
  }
}

Commit Workflow

cora commit combines review, commit message generation, and commit into a single command. It runs in two modes:

HITL mode (default)

Reviews staged changes, generates a conventional commit message, then prompts:

$ git add -A
$ cora commit
🔍 Reviewing staged changes (3 files, 247 lines)...
✅ No issues found (quality score: 9.2/10)

📝 Generated commit message:
  feat(auth): add session expiry validation

Accept commit message? [Y]es / [E]dit / [N]o › y
✅ Committed: feat(auth): add session expiry validation

Y (or Enter) — accept the message and commit
E — open $EDITOR to edit, then commit
N — abort

YOLO mode (`--yolo`)

Auto-commits with no prompts. Use for trusted workflows (e.g. CI, rapid iteration):

$ cora commit --yolo

Flags

Flag	Description
`--yolo`	Auto-commit without prompts
`--force`	Commit even if quality gate fails
`--no-review`	Skip review, only generate commit message
`--edit`	Always open `$EDITOR`
`--stream`	Stream LLM response in real-time
`-q, --quiet`	Suppress all output except the result

Quality gate integration

If the quality gate is enabled and returns FAIL, cora commit blocks the commit unless --force is passed:

$ cora commit --force

Gate every commit (agent first, hook as fallback)

Three layers, strongest first:

Agent judge. The agent reviews git diff --cached with the /cora-review skill and calls cora.review_finalize with the exact staged diff. Cora records the verdict for that change set.
Pre-commit hook (cora hook install). If the staged diff is already validated it passes; otherwise it runs cora review --staged --ci --severity major (an LLM when configured, deterministic checks otherwise). Exit 2 denies the commit; any other exit is a tooling error, reported as "not a pass".
CI. Run cora review on pull requests to develop, and fail the job when its secrets are missing.

cora hook verify prints the state of the staged diff: exit 0 validated, 10 validated but blocking, 11 not validated. Any edit to the staged change invalidates the verdict (24 h expiry).

To stop Claude Code from committing unreviewed changes or using --no-verify, add this to .claude/settings.json:

{
  "hooks": {
    "PreToolUse": [
      { "matcher": "Bash", "hooks": [{ "type": "command", "command": "cora hook claude-guard" }] }
    ]
  }
}

The guard is a safety net for honest agents, not a sandbox; the hook and CI remain the enforcement.

Configuration File

The .cora.yaml file provides persistent configuration. Place it in your project root.

File roles:

File	Purpose
`~/.cora/auth.toml`	API key (secret, chmod 600)
`~/.cora/config.yaml`	Global defaults (provider, model, etc.)
`.cora.yaml`	Per-project overrides

# .cora.yaml — example
provider: zai
model: glm-5.1

focus:
  - security
  - performance

ignore:
  files:
    - "vendor/**"
    - "*.min.js"

Environment Variables

Environment variables override configuration file settings:

Variable	Description	Required
`CORA_API_KEY`	API key for CI (overrides auth.toml)	CI only
`CORA_PROVIDER`	Override the LLM provider	No
`CORA_MODEL`	Override the model name	No
`CORA_BASE_URL`	Override the API base URL	No
`CORA_CONFIG`	Path to alternative config file	No
`CORA_TRUST_PROJECT_CONFIG`	Set to `1` to honour `provider.base_url` from a discovered project `.cora.yaml` (otherwise ignored with a warning, since it would receive your API key)	No

Security: provider.base_url must be https:// (plain http:// only for localhost, 127.0.0.1, [::1]). A base_url found only in a discovered project .cora.yaml is ignored unless CORA_TRUST_PROJECT_CONFIG=1; set it via --base-url, CORA_BASE_URL, an explicit --config file, or your global config instead.

Provider-specific keys are auto-detected: OPENAI_API_KEY, ANTHROPIC_API_KEY, GROQ_API_KEY, ZAI_API_KEY

Working with Monorepos

cora works well in monorepo setups. Use include patterns to limit review scope to specific packages:

# Review only the backend package
$ cora review --staged --include "packages/backend/**"

# Exclude test and generated files
$ cora review --staged --exclude "**/*.test.ts" --exclude "**/generated/**"

Alternatively, set include/exclude patterns in .cora.yaml for persistent configuration.

Exit Codes

cora uses standard exit codes for scripting and CI integration:

Code	Meaning	CI Behavior
`0`	No issues found	Pass
`1`	Issues found	Fail (warning/error)
`2`	Quality gate failed or review blocked	Fail (threshold exceeded)
`3`	Authentication error	Fail (missing API key)

MCP Server

cora includes a built-in MCP server for AI coding agent integration.

# Start MCP server (JSON-RPC 2.0 over stdio)
cora mcp

See Configuration → MCP Server for setup guides for Claude Code, Cursor, and other agents.

Tech Debt Tracking

cora automatically tracks review findings over time, letting you answer "is our codebase getting better or worse?"

# Show debt report
$ cora debt

# Show trend graph over review history
$ cora debt --trend

# Machine-readable output for CI dashboards
$ cora debt --json

# Shields.io badge JSON (embed in README)
$ cora debt --badge

# Estimated fix time
$ cora debt --estimate

# Filter by date or git tag
$ cora debt --since 2026-06-01
$ cora debt --since v0.4.5

# Filter by branch
$ cora debt --branch develop

Snapshots are saved to .cora/history/ after each review. Configure in .cora.yaml:

debt:
  enabled: true           # default
  history_dir: .cora/history
  retention_days: 90

Quality score ranges from 0 (worst) to 10 (best). Each finding reduces the score:

Critical: -2.0
Major: -1.0
Minor: -0.3
Info: -0.1

Trend indicators: ▼ improving (fewer findings), ► stable, ▲ worsening (more findings).

Tips

Use cora review with no flags for the fastest pre-commit feedback
Combine --format json with --base main in CI pipelines
Use cora scan . --incremental for large codebases — only changed files are analyzed
For large scans that hit provider token limits or rate-limits, lower the batch size with --batch-files 10 (default: 20). Failed batches are skipped by default — pass --no-continue-on-batch-error to abort instead
Use --quiet for minimal output and --severity to filter by severity level
Use cora auth login to store API keys securely instead of environment variables

Navigation Menu

Repository navigation

usage.md

usage.md

Usage

Review Modes

Review with your agent (no LLM setup)

Output Formats

Commit Workflow

HITL mode (default)

YOLO mode (`--yolo`)

Flags

Quality gate integration

Gate every commit (agent first, hook as fallback)

Configuration File

Environment Variables

Working with Monorepos

Exit Codes

MCP Server

Tech Debt Tracking

Tips

FilesExpand file tree

usage.md

Latest commit

History

usage.md

File metadata and controls

Usage

Review Modes

Review with your agent (no LLM setup)

Output Formats

Commit Workflow

HITL mode (default)

YOLO mode (--yolo)

Flags

Quality gate integration

Gate every commit (agent first, hook as fallback)

Configuration File

Environment Variables

Working with Monorepos

Exit Codes

MCP Server

Tech Debt Tracking

Tips

YOLO mode (`--yolo`)