richfrem/exploration-session-brief

<example> <commentary>User wants to start a new exploration session.</commentary> User: Let's kick off a new exploration session. I want to capture a brief for the problem I've been thinking about. Agent: [invokes exploration-session-brief, starts Stage 1 context gathering] </example> <example> <commentary>User has a vague idea and wants structured help thinking it through.</commentary> User: I have a rough idea to explore — somewhere between a process improvement and a new feature. Can you help me think it through and write it up? Agent: [invokes exploration-session-brief, guides user through the 3-stage co-authoring workflow] </example> <example> <commentary>Explanation request — do NOT invoke. Only trigger when user actively wants to start or capture a brief, not when asking about the skill.</commentary> User: What does the exploration-session-brief skill do? Don't start it yet. Agent: [explains the skill without invoking it] </example> <example> <commentary>Handoff requests route to exploration-handoff, not this skill.</commentary> User: I finished my exploration session, help me write the handoff package. Agent: [invokes exploration-handoff, NOT exploration-session-brief] </example>

Exploration Session Brief (Interactive Co-Authoring)

Note: This skill runs fully interactively via Claude — no script needed. execute.py is a planned batch-mode convenience wrapper that hasn't been built yet, but the core skill works now. The intake-agent provides an alternative agentic dispatch path.

This skill provides a structured, 3-stage interactive workflow for generating an Exploration Session Brief. Guide the user through each stage in sequence — do not skip ahead or dump the full brief at once.

Important Note for Agents: Do NOT passively run a bash script or dump a massive block of markdown. You must guide the user through the following 3 stages.

Stage 1: Context Gathering

Your goal is to understand the boundaries of the exploration before drafting anything. Ask all three questions together in a single message:

1. Domain: What category best fits this exploration?

   • Software feature or system — new capability, redesign, or technical spike
   • Business process — workflow, approval flow, operations improvement
   • Risk or compliance — mitigation strategy, audit finding, policy gap
   • Research or strategy — market analysis, competitive review, roadmap decision
   • Other — describe it briefly

2. Trigger: What specific event, pain point, or decision caused us to start this session right now?

3. Raw material: Do you have any notes, transcripts, screenshots, or prior docs to share? (You can brain-dump freely — messy is fine.)

Wait for the user's response. If any answer is too sparse to proceed (e.g., one-word domain, no trigger explained), ask one targeted follow-up before moving to Stage 2. Do not proceed until you have a clear trigger and at least one concrete detail.

Stage 2: Section-by-Section Refinement

Build the brief iteratively — do not write the entire document in one pass.

1. Propose the Outline: Based on the domain from Stage 1, propose a section list using the appropriate template below. Present it as a numbered list and ask the user: "Does this structure fit? Anything to add or remove?"

   | Domain | Suggested sections | |---|---| | Software feature/system | Problem Statement · Stakeholders · Current Behavior · Desired Behavior · Constraints · Open Questions | | Business process | Problem Statement · Stakeholders · Current Process · Pain Points · Desired State · Constraints · Open Questions | | Risk/compliance | Risk Description · Affected Parties · Current Exposure · Mitigation Options · Constraints · Open Questions | | Research/strategy | Research Question · Context · What We Know · What We Don't Know · Success Criteria · Open Questions |

2. Curate: Apply any changes the user requests. If they want a custom section, add it. Do not argue for the template.

3. Draft section by section: For each section, write a 2–5 sentence draft using only information from Stage 1. Present it and ask: "What should we keep, cut, or change?" Apply edits before moving to the next section. Mark anything inferred (not stated by user) as [UNCONFIRMED].

Stage 3: Reader Testing (Blind Spot Analysis)

Once all sections are drafted and approved, stress-test the brief before writing it out:

1. Pick the most likely first reader — the person who will act on this brief (e.g., an engineer building a spec, a manager approving budget, a team changing a process).
2. Predict exactly 3 questions that reader would ask after reading the brief if they had never been part of this session. Make the questions specific to this brief's content, not generic.
3. Present the 3 questions to the user: "If [reader] reads this, they'll likely ask: [Q1], [Q2], [Q3]. Does the brief answer these? Should we add answers inline or list them in ## Open Questions?"
4. Apply whatever the user decides.

Anti-Hallucination Rules

• Do NOT assume the output will be a software product unless the user says so. Ensure language remains agnostic (e.g., use "Solution" instead of "App").
• Clearly demarcate proven facts from assumptions using [CONFIRMED] and [UNCONFIRMED] markers.
• Never fake user personas or edge cases; derive them strictly from the user's Context Gathering dump.

Final Output Destination

Write the approved, refined markdown content to: exploration/session-brief.md

Use the Canonical Session Brief Schema headers defined in agents/intake-agent.md so that exploration-workflow Block 1 can silently read and hydrate the dashboard without re-asking setup questions.