tsh-task-extracting - SKILL.md Agent Skill

name: tsh-task-extracting description: Identify and structure epics and user stories from workshop materials (cleaned transcripts, Figma designs, codebase analysis, and other documents). Produces a business-oriented task breakdown with dependencies, assumptions, and open questions. user-invocable: false

Task Extraction

This skill helps you identify discrete pieces of work (epics and user stories) from discovery workshop materials and structure them into a clear, business-oriented task breakdown. The output is intended for stakeholder review and eventual Jira creation — it is NOT a technical specification or implementation plan.

What This Skill Produces

Intent brief: A short business-first scope brief that is approved before extraction begins
Epics: High-level work streams with business descriptions and success criteria
User Stories: Discrete deliverables in "As a… I want… So that…" format with source traceability and scenario-based acceptance criteria
Dependencies: Relationships between epics and stories
Assumptions & Open Questions: Gaps that need stakeholder input

What This Skill Does NOT Produce

Technical architecture or implementation details (use tsh-architect agent for that)
Detailed acceptance criteria from a QA perspective (use tsh-context-engineer agent for that)
Story point estimates (left for team estimation sessions)
Sprint or release planning

Task Extraction Process

Use the checklist below and track your progress:

Extraction progress:
- [ ] Step 1: Gather and review all input materials
- [ ] Step 2: Draft the intent brief
- [ ] Step 3: Review Gate 0 with the user and approve the intent brief
- [ ] Step 4: Identify high-level work streams (epics)
- [ ] Step 5: Break down epics into user stories
- [ ] Step 6: Write business-oriented descriptions with source traceability
- [ ] Step 7: Map dependencies between tasks
- [ ] Step 8: Identify assumptions and out-of-scope items
- [ ] Step 9: Flag ambiguities and ask clarifying questions
- [ ] Step 10: Present task list for user validation
- [ ] Step 11: Save the intent brief and extracted tasks documents

Step 1: Gather and review all input materials

Collect and thoroughly review all available workshop materials:

workshop-context-summary.md (if available): Explore Mode summary that captures business context, likely epics, and ambiguities before extraction
Cleaned transcript (cleaned-transcript.md): Primary source — review all discussion topics, decisions, action items, and open questions
Figma/FigJam designs: If available, analyze screens, flows, and annotations for functional requirements
Existing codebase: Use tsh-codebase-analysing skill to understand what already exists and what needs to be built
Other documents: Confluence pages, shared documents, email threads, or any other reference materials provided by the user
PDF documents: If available, read PDF files using the pdf-reader tool to extract requirements, process descriptions, business rules, or any other relevant content provided by the client
Project baseline (specifications/projects/<project-name>/task-baseline.md, if present): Review for continuity, overlap, and existing backlog context

Create a mental model of the full scope discussed during the workshop before proceeding to extraction.

Step 2: Draft the intent brief

Before extracting tasks, draft an intent-brief.md that captures the business intent of the workshop. Keep it concise and focused on scope decisions, not backlog detail.

Include at least:

Goal
In scope
Out of scope
Key stakeholders / actors
Likely epic candidates
Baseline overlap or existing backlog notes
Open questions

If the user requested Explore Mode, this draft may be accompanied by a workshop-context-summary.md, but it still must not become a backlog artifact.

Step 3: Review Gate 0 with the user and approve the intent brief

Present the intent brief to the user for validation before backlog extraction begins. Ask whether the scope, exclusions, and candidate epics are correct. Iterate until the user approves.

This is Gate 0. Do not proceed to extraction until the user approves the intent brief.

Step 4: Identify high-level work streams (epics)

From the gathered materials, identify distinct work streams that represent major deliverables or feature areas:

Look for natural groupings of related functionality
Each epic should represent a cohesive business capability (e.g., "User Authentication", "Payment Processing", "Reporting Dashboard")
Epics should be independent enough to be delivered and validated separately where possible
Aim for 3-10 epics per workshop (if you find more, some may be too granular; if fewer, some may be too broad)

For each epic, draft:

A clear, business-oriented title
A 2-3 sentence business description explaining the value
High-level success criteria (what "done" looks like from a business perspective)

Step 5: Break down epics into user stories

For each epic, identify the individual user stories that compose it:

Each story should represent a single, deliverable piece of user-facing functionality
Stories should be small enough to be completed in a single sprint (as a guideline)
Look for stories in: feature descriptions, user workflows, business rules, data requirements, integration points
Include non-functional stories where explicitly discussed (e.g., "As an admin, I want audit logging so that compliance requirements are met")

Step 6: Write business-oriented descriptions with source traceability

For each user story, write:

Title: Short, descriptive, action-oriented (e.g., "User can reset password via email link")
User story: "As a [role], I want [capability] so that [benefit]"
Source: A concise traceability field that points back to the workshop material(s), baseline entry, or context summary used to derive the story
Acceptance criteria: Checklist of verifiable business conditions, written primarily as concise GIVEN / WHEN / THEN scenarios
High-level technical notes: Brief notes ONLY where the workshop discussion explicitly mentioned technical considerations (e.g., "Discussed using SSO integration", "Needs to support 10k concurrent users"). Do NOT invent technical details
Priority suggestion: Based on discussion emphasis and dependencies (Critical / High / Medium / Low)
Additional acceptance checks: Optional non-scenario checks when a GIVEN / WHEN / THEN statement is not sufficient

Important: Keep descriptions in business language. Avoid implementation jargon. The goal is for any stakeholder to understand what will be delivered without technical knowledge.

Step 7: Map dependencies between tasks

Identify relationships between epics and stories:

Blocked by: Story A cannot start until Story B is complete
Related to: Stories that share context but don't block each other
Epic dependencies: When one epic must be delivered before another can begin

Use clear notation (e.g., "Story 1.2 is blocked by Story 1.1") in the dependencies section.

Step 8: Identify assumptions and out-of-scope items

Document:

Assumptions: Decisions you made based on interpretation of the materials where the intent was not 100% clear. Label each assumption clearly so stakeholders can confirm or correct it.
Out of scope: Items that were explicitly excluded during the workshop, or that you identified as beyond the current discussion scope. This prevents scope creep and sets clear boundaries.

Step 9: Flag ambiguities and ask clarifying questions

Review all extracted tasks and identify:

Stories where you are not confident about the scope or intent
Conflicting information between different materials (e.g., transcript says one thing, Figma shows another)
Missing information that would significantly affect the task breakdown

Use askQuestions to clarify these items with the user. Ask exactly one question per askQuestions call. Each question must clearly identify the specific epic or story it relates to — include the story identifier and title in the question header and context (e.g., "[Epic: User Auth > Story 1.2: User can log in] The transcript mentions SSO but the Figma shows email/password only. Which scope is correct?"). This ensures each popup is self-contained and the user can focus on one decision at a time.

Step 10: Present task list for user validation

Present each story to the user individually for validation using one askQuestions call per story. Each question should include the story's full context: parent epic title, story title, and a brief summary of the acceptance criteria. Ask: "Is this story correct? Should it be split, merged, modified, or removed?"

After presenting all stories, ask one final workflow-level question: "Did I miss any tasks that should be added?"

Iterate based on feedback until the user approves the task list.

This is Review Gate 1 — the user must approve the task list before proceeding to Jira formatting.

Step 11: Save the intent brief and extracted tasks documents

Generate the final outputs following the ./intent-brief.example.md and ./extracted-tasks.example.md templates.

Save intent-brief.md and extracted-tasks.md to specifications/<workshop-name>/.

The extracted tasks document must preserve source traceability in each story and keep acceptance criteria in business-friendly scenario form.

Connected Skills

tsh-transcript-processing - provides the cleaned transcript used as primary input
tsh-codebase-analysing - for understanding existing system context when analyzing scope
tsh-task-analysing - for business-context exploration, baseline comparison, and intent brief drafting before extraction