Write Tickets

Write Tickets

Overview

Transform unstructured inputs (meeting notes, Slack threads, rough ideas) into well-structured project tickets. The workflow runs three steps in sequence, each in its own subagent. Data is passed between steps via temp files.

Pipeline

[Subagent 1: Research]  →  /tmp/write-tickets-research.md
        ↓
[Subagent 2: Write]     →  /tmp/ticket-N-*.md + /tmp/write-tickets-manifest.json
        ↓
[Subagent 3: Publish]   →  Tickets/issues created (Jira and/or GitHub per Target), temp files cleaned up

How to Run

Launch each step as a subagent in sequence. Each subagent reads its step file and follows it. Pass the handoff file paths in the subagent prompt so it knows where to read/write.

Skipping steps: If the user already has gathered data → start at Step 2. If they already have finalized ticket markdown in temp files → start at Step 3.

Step 1: Research

Launch a subagent with this prompt:

Read and follow the skill step file at: <absolute-path-to>/steps/research.md

When research is complete, write the gathered data to /tmp/write-tickets-research.md
using this format:

  ## Requirements
  <the user's requirements>

  ## Research Findings
  <codebase and/or online research results, or "None">

  ## Source Links
  <preserved URLs from user input, or "None">

Return a summary of what was gathered.

When the subagent returns, proceed to Step 2.

Step 2: Write

Launch a subagent with this prompt:

Read and follow the skill step file at: <absolute-path-to>/steps/write.md

Read the gathered data from /tmp/write-tickets-research.md — this contains
the requirements, research findings, and source links from Step 1.

The ticket templates are at: <absolute-path-to>/templates/

After all tickets are finalized, write a manifest to /tmp/write-tickets-manifest.json:
{
  "tickets": [
    {"number": 1, "title": "...", "type": "Story|Bug|Task", "tempFile": "/tmp/ticket-1-slug.md"},
    ...
  ]
}

Return the ticket plan summary and manifest path.

When the subagent returns, proceed to Step 3.

Step 3: Publish

Before launching the publish subagent: Resolve publish target so you use the correct step and settings.

1. Resolve .publish-settings.md: Read <workspace-root>/.publish-settings.md, or if missing ~/.publish-settings.md. If neither exists, use AskQuestion to ask the user where to create it (project vs home) or to specify target manually; then create from templates/publish-settings.md or note the chosen target.
2. Parse the Target field: jira, github, or jira,github (comma-separated, either order). This decides which publish step(s) to run and which section of the config each step uses.

Then launch subagent(s) according to Target:

• Target is jira only: One subagent with steps/publish-jira.md and the publish-settings template path. It uses the Jira section of .publish-settings.md.
• Target is github only: One subagent with steps/publish-github.md and the publish-settings template path. It uses the GitHub section of .publish-settings.md.
• Target is jira,github (or github,jira): Two subagents in sequence. First: run steps/publish-jira.md with publish-settings; tell it do not clean up temp files (the GitHub step will run next and will clean up). Second: run steps/publish-github.md with publish-settings; it cleans up all temp files at the end.

Use this prompt shape (replace STEP_FILE and optional cleanup note as above):

Read and follow the skill step file at: <absolute-path-to>/steps/STEP_FILE

Read the ticket manifest from /tmp/write-tickets-manifest.json to get the list
of finalized tickets and their temp file paths.

The publish-settings template is at: <absolute-path-to>/templates/publish-settings.md
[If Target is jira,github and this is the Jira step: Do not delete any temp files; the GitHub publish step will run next and will clean up.]

After all tickets/issues are created, clean up all temp files (unless instructed above to skip):
- /tmp/write-tickets-research.md
- /tmp/write-tickets-manifest.json
- All /tmp/ticket-N-*.md files

Return the list of created ticket keys or issue links.

Shared Conventions

• Publish settings: A single config file .publish-settings.md controls where tickets go and how to connect. Lookup order: project root (<workspace>/.publish-settings.md), then user home (~/.publish-settings.md). The file has a Target field: jira, github, or jira,github. When Target includes jira, the Jira section (project key, component, sprint, base URL, issue type mapping) is used by steps/publish-jira.md. When Target includes github, the GitHub section (optional type→labels, optional repo) is used by steps/publish-github.md. If no file exists, the user is prompted to create one or specify manually. Template: templates/publish-settings.md.
• Ticket document content is produced in the write step using this skill's templates/ (feature, bug, tech-debt, spike) and the ticket writing rules in steps/write.md.
• Jira creation is in steps/publish-jira.md (MCP or acli); GitHub Issues creation is in steps/publish-github.md (GitHub CLI). Never use a separate jira-expert subagent for creation.
• Links: When showing a ticket key, always use a link: [RETIRE-1234](https://gustohq.atlassian.net/browse/RETIRE-1234).
• No local file paths in ticket content. Use GitHub links only, using the current project's repository URL (e.g. from git remote get-url origin or the workspace context).

Handoff Files

All temp files are cleaned up by Step 3 after tickets/issues are created (by the last publish step when Target is both).

Steps Reference