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:
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
Trigger: What specific event, pain point, or decision caused us to start this session right now?
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.
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 |
Curate: Apply any changes the user requests. If they want a custom section, add it. Do not argue for the template.
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:
- 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).
- 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.
- 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?"
- 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/sessions/session-brief.md (or a timestamped equivalent).
1---2name: exploration-session-brief3description: Interactive co-authoring skill for the wide end of the exploration funnel. Captures and refines the core intent, whether the outcome is a software app, a business process improvement, research analysis, or strategic roadmap. Guides users through gathering context, iteratively drafting the brief, and testing for blind spots.4---5
6<example>
7<commentary>User wants to start a new exploration session.</commentary>
8User: Let's kick off a new exploration session. I want to capture a brief for the problem I've been thinking about.
9Agent: [invokes exploration-session-brief, starts Stage 1 context gathering]
10</example>
11
12<example>
13<commentary>User has a vague idea and wants structured help thinking it through.</commentary>
14User: 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?
15Agent: [invokes exploration-session-brief, guides user through the 3-stage co-authoring workflow]
16</example>
17
18<example>
19<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>
20User: What does the exploration-session-brief skill do? Don't start it yet.
21Agent: [explains the skill without invoking it]
22</example>
23
24<example>
25<commentary>Handoff requests route to exploration-handoff, not this skill.</commentary>
26User: I finished my exploration session, help me write the handoff package.
27Agent: [invokes exploration-handoff, NOT exploration-session-brief]
28</example>
29
30# Exploration Session Brief (Interactive Co-Authoring)
31
32> **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](../../agents/intake-agent.md) provides an alternative agentic dispatch path.
33
34This 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.
35
36**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.
37
38## Stage 1: Context Gathering
39Your goal is to understand the boundaries of the exploration *before* drafting anything. Ask all three questions together in a single message:
40
411. **Domain:** What category best fits this exploration?
42 - *Software feature or system* — new capability, redesign, or technical spike
43 - *Business process* — workflow, approval flow, operations improvement
44 - *Risk or compliance* — mitigation strategy, audit finding, policy gap
45 - *Research or strategy* — market analysis, competitive review, roadmap decision
46 - *Other* — describe it briefly
47
482. **Trigger:** What specific event, pain point, or decision caused us to start this session right now?
49
503. **Raw material:** Do you have any notes, transcripts, screenshots, or prior docs to share? *(You can brain-dump freely — messy is fine.)*
51
52Wait 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.
53
54## Stage 2: Section-by-Section Refinement
55Build the brief iteratively — do not write the entire document in one pass.
56
571. **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?"*
58
59 | Domain | Suggested sections |
60 |---|---|
61 | Software feature/system | Problem Statement · Stakeholders · Current Behavior · Desired Behavior · Constraints · Open Questions |
62 | Business process | Problem Statement · Stakeholders · Current Process · Pain Points · Desired State · Constraints · Open Questions |
63 | Risk/compliance | Risk Description · Affected Parties · Current Exposure · Mitigation Options · Constraints · Open Questions |
64 | Research/strategy | Research Question · Context · What We Know · What We Don't Know · Success Criteria · Open Questions |
65
662. **Curate:** Apply any changes the user requests. If they want a custom section, add it. Do not argue for the template.
67
683. **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]`.
69
70## Stage 3: Reader Testing (Blind Spot Analysis)
71Once all sections are drafted and approved, stress-test the brief before writing it out:
72
731. 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).
742. 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.
753. 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`?"*
764. Apply whatever the user decides.
77
78## Anti-Hallucination Rules
79- 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").
80- Clearly demarcate proven facts from assumptions using `[CONFIRMED]` and `[UNCONFIRMED]` markers.
81- Never fake user personas or edge cases; derive them strictly from the user's Context Gathering dump.
82
83## Final Output Destination
84Write the approved, refined markdown content to: `exploration/sessions/session-brief.md` (or a timestamped equivalent).
aibot88