Pipeline Summary

Pipeline Summary

Create a GitHub issue documenting bugs encountered during a pipeline run and a PR from the feature branch into the target branch.

Arguments

/autoskillit:pipeline-summary {bug_report_path} {feature_branch} {target_branch} {workspace} [{token_summary_path}] [{closing_issue}]

• bug_report_path — Path to the JSON file containing bug metadata
• feature_branch — Name of the branch containing all accumulated fixes
• target_branch — Branch to create the PR against (e.g., "main")
• workspace — Path to the git repository workspace
• token_summary_path — (Optional) Path to a JSON file with token/timing data written by the orchestrator. When absent or the file does not exist, the skill operates exactly as today with no token table in the PR body.
• closing_issue — (Optional) GitHub issue number whose ## Requirements section should be extracted and embedded in the PR body. When absent or empty, requirements extraction is skipped.

When to Use

• End of a pipeline run with collect_on_branch enabled
• Any pipeline that accumulates fixes on a feature branch and needs a summary

Critical Constraints

NEVER:

• Fail the pipeline if gh is not available or not authenticated — write a local summary instead
• Create empty issues or PRs (skip if no bugs to report)
• Modify any source code — this skill only creates GitHub artifacts and a summary file

ALWAYS:

• Check gh auth status before attempting GitHub operations
• Push the feature branch before creating the PR
• Write a local summary markdown file regardless of GitHub availability
• Output summary_path=<path> for capture by the orchestrator
• If GitHub operations succeed, also output issue_url=<url> and pr_url=<url>

Workflow

Step 1: Parse Arguments

Parse up to six positional arguments from the prompt. The fifth (token_summary_path) and sixth (closing_issue) are optional.

Step 2: Read Bug Report

Read the JSON file at {bug_report_path}. Expected structure:

[
  {
    "step": "string — pipeline step where failure occurred",
    "error": "string — error description",
    "fix": "string — what was done to fix it",
    "iteration": "number — which bugfix iteration"
  }
]

If the file is empty, contains [], or doesn't exist, write a clean-run summary and exit successfully.

Step 3: Write Local Summary

Write a markdown summary to {workspace}/run-summary.md:

• Title: "Pipeline Run Summary — {date}"
• Bug count and fix count
• Table of all bugs with step, error, fix, iteration
• Branch info: feature branch name, target branch

Output: summary_path={workspace}/run-summary.md

Step 3b: Append Token+Timing Table (if token_summary_path provided)

If token_summary_path was provided as a fifth argument and the file exists, read it as JSON. The JSON has the structure:

{
  "steps": [
    {
      "step_name": "string",
      "input_tokens": 0,
      "output_tokens": 0,
      "cache_creation_input_tokens": 0,
      "cache_read_input_tokens": 0,
      "invocation_count": 0,
      "elapsed_seconds": 0.0
    }
  ],
  "total": {
    "input_tokens": 0,
    "output_tokens": 0,
    "cache_creation_input_tokens": 0,
    "cache_read_input_tokens": 0,
    "total_elapsed_seconds": 0.0
  }
}

Append the following two markdown sections to run-summary.md:

## Token Usage

| Step | Input | Output | Cache Write | Cache Read | Calls | Elapsed (s) |
|------|-------|--------|-------------|------------|-------|-------------|
| {step_name} | {input_tokens} | {output_tokens} | {cache_creation_input_tokens} | {cache_read_input_tokens} | {invocation_count} | {elapsed_seconds:.1f} |
| **Total** | {total.input_tokens} | {total.output_tokens} | {total.cache_creation_input_tokens} | {total.cache_read_input_tokens} | — | {total.total_elapsed_seconds:.1f} |

If token_summary_path is absent or the file does not exist, skip this step — no token table is added.

Step 4: Check GitHub Availability

Run gh auth status 2>/dev/null. If exit code is non-zero or gh is not found:

• Log "GitHub CLI not available or not authenticated — skipping issue/PR creation"
• Exit successfully (the local summary is sufficient)

Step 5: Push Feature Branch

cd {workspace}
git push -u origin {feature_branch}

If push fails (no remote, network issue), log the error and exit successfully.

Step 5b: Fetch Requirements from Closing Issue (if closing_issue known)

• If closing_issue was provided as the sixth argument:

  gh issue view {closing_issue} --json body -q .body
  

  Extract the ## Requirements section: requirements_section = everything from ## Requirements to the next ## heading or end of body, whichever comes first.
• If gh auth is unavailable or closing_issue is not provided: skip gracefully — requirements_section = "".

Step 6: Create GitHub Issue

Write the issue body to a temp file, then:

TEMP_ISSUE_BODY="temp/pipeline-summary/issue_body_$(date +%Y%m%d-%H%M%S).md"
mkdir -p "$(dirname "${TEMP_ISSUE_BODY}")"
# [write the issue body content to ${TEMP_ISSUE_BODY} here]
gh issue create \
  --title "Pipeline Run Summary — {date}: {bug_count} bug(s) fixed" \
  --body-file "${TEMP_ISSUE_BODY}" \
  --label "pipeline-summary"

Capture the issue URL from stdout. If the label doesn't exist, retry without --label.

Output: issue_url={url}

Step 7: Create Pull Request

Write the PR body to a temp file (reference the issue), then:

TEMP_PR_BODY="temp/pipeline-summary/pr_body_$(date +%Y%m%d-%H%M%S).md"
mkdir -p "$(dirname "${TEMP_PR_BODY}")"
# [write the PR body content to ${TEMP_PR_BODY} here]
gh pr create \
  --title "Pipeline fixes — {date}" \
  --body-file "${TEMP_PR_BODY}" \
  --base {target_branch} \
  --head {feature_branch}

The PR body (temp_pr_body) contains:

• ## Summary — bug count and branch info
• ## Requirements (if requirements_section is non-empty from Step 5b)
• Closes #{closing_issue} (if closing_issue was provided)
• Bug table from Step 3
• Token/timing table from Step 3b (if available)

Capture the PR URL from stdout.

Output: pr_url={url}

Output

• Always: summary_path={workspace}/run-summary.md
• If GitHub available: issue_url={url} and pr_url={url}

Orchestrator Calling Convention

For recipe authors who want to include token/timing data in the PR body:

1. Call the get_token_summary MCP tool to retrieve current pipeline token data.
2. Write the JSON result to temp/token_summary_{timestamp}.json using a run_python step. (relative to the current working directory) The run_python step executes in the MCP server process and has access to the live ToolContext via the server context; call ctx.token_log.get_report() and ctx.token_log.compute_total(), then write {"steps": ..., "total": ...} as JSON. Capture the output path via print(f"token_summary_path={out}").
3. Pass the file path as the fifth positional argument to run_skill pipeline-summary.

Note: The headless session for pipeline-summary runs in a separate process with its own (empty) token log, which is why the file-based handoff is required. run_python steps share the live in-process token log with the MCP server, so they can access accumulated timing data directly without a network call.