| crash_report | JSON string containing a Unity/Backtrace crash report object or an array of crash report objects. Ignored when crash_report_file is provided. | no | "" |
| crash_report_file | Local path or remote URL (http/https) to a JSON file containing crash report data (object or array). Takes priority over crash_report when provided. Use this for large payloads that exceed GitHub Actions input size limits. | no | "" |
| anthropic_api_key | Anthropic API key for Claude | yes | — |
| github_token | GitHub token for creating PRs | no | ${{ github.token }} |
| base_branch | Base branch for the PR (defaults to repository default branch) | no | — |
| branch_prefix | Prefix for the fix branch name | no | fix/crash- |
| max_turns | Maximum number of turns for Claude Code CLI | no | 100 |
| allowed_tools | Comma-separated list of tools allowed for Claude | no | Read,Edit,Write,Bash,Grep,Glob,WebSearch,WebFetch |
| blocked_directories | Comma-separated repo-relative directories where Claude changes are forbidden (e.g. "Assets/Core,Packages") | no | "" |
| pr_labels | Comma-separated labels for the PR | no | crash-fix,auto-generated |
| dry_run | If true, only analyze the crash report without creating a PR | no | false |
| use_router | If true, route the crash report to remote workflow(s) based on router rules and skip local Claude execution when matched | no | false |
| router_rules_json | JSON array of router rules. Each rule includes match_prefixes, target, and optional rewrite | no | "" |
| router_mode | Routing mode: first-match (single dispatch) or fanout (dispatch all matches) | no | first-match |
| router_default_target_json | Optional JSON target used when no router rule matches | no | "" |
| fix_strategy | Fix strategy for Claude: "minimal" applies the smallest targeted change to stop the crash; "refactor" allows broader refactoring of the affected area; "aggressive" gives Claude full freedom to improve the code, fix similar latent issues nearby, and restructure patterns | no | minimal |
| additional_prompt | Additional instructions appended to the default prompt (project context, special rules, etc.) | no | "" |
| claude_permission_mode | Claude permission mode (plan, dontAsk, etc.) | no | plan |
| claude_auto_apply_plan | If true and permission mode is plan, run a second pass in dontAsk mode to apply changes | no | true |
| claude_model | Claude model to use | no | claude-opus-4-6 |
| claude_effort | Reasoning effort level (low, medium, high) | no | high |
| claude_betas | Comma-separated Claude beta flags | no | "" |