| mode | Action mode: auto-fix, command, manual, cleanup | yes | — |
| working-directory | Directory where Claude Code runs | no | . |
| max-turns | Max Claude conversation turns per attempt. Use -1 for unlimited. | no | 50 |
| max-retries | Max retry attempts (each gets fresh max-turns budget) | no | 3 |
| max-retry-commands | Max /ci-assistant retry commands per PR. Use -1 for unlimited. | no | 2 |
| max-alternative-commands | Max /ci-assistant alternative commands per PR. Use -1 for unlimited. | no | 3 |
| max-suggest-commands | Max /ci-assistant suggest commands per PR. Use -1 for unlimited. | no | 3 |
| max-explain-commands | Max /ci-assistant explain commands per PR. Use -1 for unlimited. | no | 3 |
| max-total-commands | Max total non-admin commands per PR. Use -1 for unlimited. | no | 20 |
| model | Claude model to use (e.g. 'claude-sonnet-4-6', 'claude-opus-4-6', 'claude-haiku-4-5-20251001') | no | claude-sonnet-4-6 |
| skip-permissions | Pass --dangerously-skip-permissions to Claude Code. Safe in CI since the runner is ephemeral. Set to false to require permission checks. | no | true |
| allowed-tools | Comma-separated list of tools Claude Code can use (e.g. 'Bash,Edit,Read,Write,WebSearch'). Empty means all tools available. | no | "" |
| disallowed-tools | Comma-separated list of tools Claude Code cannot use (e.g. 'WebSearch,WebFetch'). Takes precedence over allowed-tools. | no | "" |
| append-system-prompt | Additional system prompt appended to Claude Code's default system prompt. Use for project-specific instructions not covered by CLAUDE.md. | no | "" |
| admin-users | Comma-separated GitHub usernames of admin users (e.g. user1,user2) | no | "" |
| banned-users | Comma-separated GitHub usernames permanently banned from using CI Assistant | no | "" |
| slack-failure-channel | Slack channel ID for fix notifications (optional, everything works without Slack) | no | "" |
| slack-thread-ts | Slack thread timestamp to post as a reply (optional) | no | "" |
| slack-bot-token | Slack Bot OAuth Token (optional) | no | "" |
| failed-run-id | The workflow run ID that failed | no | "" |
| failed-branch | The branch that failed | no | "" |
| failed-sha | The commit SHA that failed | no | "" |
| failed-pr-number | The PR number (if failure was on a PR) | no | "" |
| comment-pr-number | The PR number the comment was made on (command mode) | no | "" |
| auto-fix-prompt | Prompt template for automatic fix attempts. Placeholders: {{FAILURE_LOGS}}, {{REPO}}, {{BRANCH}}, {{SHA}}, {{PREVIOUS_SUGGESTIONS}}, {{USER_CONTEXT}}, {{CONVERSATION_HISTORY}} | no | "" |
| retry-prompt | Prompt template for retry attempts. Placeholders: {{FAILURE_LOGS}}, {{PREVIOUS_ATTEMPTS}}, {{REPRODUCTION_OUTPUT}}, {{PREVIOUS_SUGGESTIONS}}, {{USER_CONTEXT}}, {{CONVERSATION_HISTORY}}, {{REPO}}, {{BRANCH}}, {{SHA}} | no | "" |
| alternative-prompt | Prompt template for alternative suggestions. Placeholders: {{FAILURE_LOGS}}, {{PREVIOUS_SUGGESTIONS}}, {{REPO}}, {{BRANCH}}, {{SHA}}, {{CONVERSATION_HISTORY}} | no | "" |
| suggest-prompt | Prompt template for suggest commands. Placeholders: {{USER_CONTEXT}}, {{FAILURE_LOGS}}, {{CONVERSATION_HISTORY}}, {{REPO}}, {{BRANCH}}, {{SHA}} | no | "" |
| explain-prompt | Prompt template for explain commands. Placeholders: {{USER_PROMPT}}, {{FAILURE_LOGS}}, {{LATEST_FIX_DIFF}}, {{CONVERSATION_HISTORY}}, {{REPO}}, {{BRANCH}} | no | "" |
| confidence-prompt | Prompt appended to fix prompts for confidence analysis. No placeholder substitution (appended as-is). | no | "" |
| summary-prompt | Prompt appended to fix prompts for fix summary generation. Claude outputs FIX_TITLE and FIX_DESCRIPTION markers. No placeholder substitution. | no | "" |
| github-token | GitHub token for API calls (supports default github.token or GitHub App tokens) | no | "" |
| app-slug | GitHub App slug (from actions/create-github-app-token output). Used to determine the bot identity. If empty, falls back to API detection. | no | "" |
| claude-code-oauth-token | Claude Code OAuth token (from claude setup-token, uses subscription quota) | no | "" |
| anthropic-api-key | Anthropic API key (pay-per-use fallback) | no | "" |
| comment-body | The PR comment text (command mode) | no | "" |