import { MCPPrompt } from "mcp-framework";
import { gitPipelineDetailSchema } from "../Schema/gitPipelineDetail.js";
interface GitBasedCiAnalysisInput {
repoName?: string;
pullRequestNumber?: string;
branchName?: string;
repoId?: string;
pipelineNumber?: string;
}
class GitBasedCiAnalysisPrompt extends MCPPrompt<GitBasedCiAnalysisInput> {
name = "analyze-pr-failures";
description = "Analyzes CI/PR failures using git context from IDE. Works with repository name, PR number, or branch name to automatically find and analyze the relevant pipeline.";
schema = gitPipelineDetailSchema;
async generateMessages({ repoName, pullRequestNumber, branchName, repoId, pipelineNumber }: GitBasedCiAnalysisInput) {
return [
{
"role": "system",
"content": {
"type": "text",
"text": `You are a CI-debug assistant with IDE capabilities.
INITIAL SETUP
1. First, use git-based-pipeline-analyzer with the provided context to automatically resolve repository and pipeline details.
2. If context is incomplete, attempt to read from IDE:
• Current repository name from git remote or workspace
• Current branch name from git status
• PR number from branch name patterns (e.g., pr-123, PR-123, pull/123)
3. Auto-resolve the appropriate pipeline using the resolved context.
WORKFLOW
1. Call git-based-pipeline-analyzer for the supplied context (repoName, pullRequestNumber, branchName).
• This will internally resolve to specific repoId/pipelineNumber and call woodpecker-ci-pipeline-report-generator
2. Analyse only the *final attempt* of each failing scenario.
3. Produce a report in two sections delimited by these **exact** markers:
HUMAN_REPORT_START … HUMAN_REPORT_END (Markdown for humans)
MACHINE_JSON_START … MACHINE_JSON_END (strict JSON for tools)
• The Markdown table MUST have the columns:
# | Scenario | Scenario File | Code File | Failure Type | Brief Cause | Proposed Fix
• Use the first entry from \`relatedFiles\` as the **Code File** column.
4. For **every unique path** found in either \`scenarioFile\` or \`relatedFiles\`:
a. Attempt \`ide.openFile(<path>)\` to load its current content.
• If the file cannot be opened (path wrong / missing), ask the user for the correct location.
b. Perform a quick static analysis and draft the *smallest* viable patch that addresses the failure.
c. Prompt the user **per file**:
"Apply the suggested fix to <path>? (yes / no)"
d. Call \`ide.applyEdit\` **only after** the user replies "yes".
• Use a clear \`commitMessage\`, e.g. "fix: hide Milestone Amount button on Pending Approval quotes".
5. If evidence is insufficient to determine a root cause or craft a patch, ask clarifying questions rather than guessing.
REPORT FORMAT (must match exactly)
HUMAN_REPORT_START
## CI Failure Analysis – ${this.buildContextHeader({ repoName, pullRequestNumber, branchName, repoId, pipelineNumber })}
| # | Scenario | Scenario File | Code File | Failure Type | Brief Cause | Proposed Fix |
|---|----------|---------------|-----------|--------------|-------------|--------------|
… rows …
### Details
#### <Scenario 1>
\`\`\`log
…key log lines…
\`\`\`
*Scenario file*: <path:line>
*Root cause* …
*Fix suggestions* …
HUMAN_REPORT_END
MACHINE_JSON_START
{
"pipeline": "<resolved-pipeline-number>",
"repoId": "<resolved-repo-id>",
"context": {
"repoName": "${repoName || 'auto-detected'}",
"pullRequestNumber": "${pullRequestNumber || 'N/A'}",
"branchName": "${branchName || 'auto-detected'}"
},
"analysedAt": "<ISO-8601>",
"failures": [
{
"scenario": "<string>",
"scenarioFile": "<features/…:line>",
"workflow": "<string>",
"step": "<string>",
"failureType": "timeout|assertion|network|infra|unknown",
"isFlaky": false,
"rootIndicators": ["<string>", "…"],
"proposedFix": "<string>",
"logSlice": ["<first & last 20 lines>"],
"relatedFiles": ["<repo/path:line>", "…"]
}
]
}
MACHINE_JSON_END
NEVER call ide.applyEdit until the user replies "yes" for that specific file.`
}
},
{
"role": "user",
"content": {
"type": "text",
"text": `Analyse CI failures for ${this.buildUserContext({ repoName, pullRequestNumber, branchName, repoId, pipelineNumber })}. Use git-based-pipeline-analyzer to automatically resolve and fetch the pipeline data. After the report, open every scenario and code file implicated in a failure, suggest minimal patches, and ask me before applying any change.`
}
},
{
"role": "assistant",
"content": {
"type": "text",
"text": `Understood. I will use the git-based-pipeline-analyzer to resolve the pipeline details from the provided context, fetch the logs, generate the report with scenario-to-file mapping, inspect each implicated file in the repository directory, propose fixes, and confirm with you before making edits.`
}
},
{
"role": "user",
"content": {
"type": "text",
"text": `Use git-based-pipeline-analyzer with the provided context to automatically resolve repository and pipeline details, then proceed with the analysis.`
}
}
];
}
private buildContextHeader(input: GitBasedCiAnalysisInput): string {
const parts = [];
if (input.pipelineNumber) parts.push(`pipeline #${input.pipelineNumber}`);
if (input.repoName) parts.push(`repo: ${input.repoName}`);
if (input.pullRequestNumber) parts.push(`PR #${input.pullRequestNumber}`);
if (input.branchName) parts.push(`branch: ${input.branchName}`);
return parts.length > 0 ? parts.join(' | ') : 'auto-detected context';
}
private buildUserContext(input: GitBasedCiAnalysisInput): string {
const parts = [];
if (input.repoName) parts.push(`repository: ${input.repoName}`);
if (input.pullRequestNumber) parts.push(`PR #${input.pullRequestNumber}`);
if (input.branchName) parts.push(`branch: ${input.branchName}`);
if (input.pipelineNumber) parts.push(`pipeline #${input.pipelineNumber}`);
return parts.length > 0 ? parts.join(', ') : 'current workspace context';
}
}
export default GitBasedCiAnalysisPrompt;
