first-task-mapper

What This Skill Does

Converts a vague work request into a clear, safe execution map. For each field in the map, it labels the confidence level so the employee can distinguish verified facts from missing information. Finishes with a readiness decision and a recommended next action. The skill plans only — it never executes the mapped task.

Process

Ask clarifying questions via AskUserQuestion (with drafted, plausible options plus a free-text "Other") to fill in each field. Label every important field as one of:

• Confirmed — The employee explicitly confirmed this
• Assumption — Reasonable inference, not yet confirmed
• Unknown — Cannot be determined from available information
• Not confirmed — Exists but not yet verified with the employee

If $ARGUMENTS is provided, use it as the initial request. Otherwise ask the employee to describe their request.

Step 1: Restate & Confirm

Restate your understanding of the request back to the employee. Confirm via AskUserQuestion: "Is this correct?" with options "Yes, that's correct", "No, let me clarify", and "Partially — some details need fixing". If not fully confirmed, clarify before proceeding.

Step 2: Identify Business Goal

Ask via AskUserQuestion: "What is the real goal here — what problem are you trying to solve?" Draft 2-4 plausible goals grounded in the restated request from Step 1 (e.g. for a report request: "Track performance over time", "Support a specific decision", "Meet a compliance/reporting requirement", "Share status with stakeholders") — the tool always adds a free-text "Other" option too.

Label: Confirmed / Assumption / Unknown

Step 3: Define Expected Result & Output Format

Ask via AskUserQuestion: "What should the final result look like?" with options "A report or document", "A list or table", "A file (specify format)", and "An email or message".

Label: Confirmed / Assumption / Unknown

Step 4: Identify Source of Truth

Check project files for hints first (CLAUDE.md, configs) — do not guess, but use any hint found as the first, recommended option. Ask via AskUserQuestion: "Where does the data live?" with options "A database", "A spreadsheet", "A file", and "A person or team".

Label: Confirmed / Assumption / Unknown

Step 5: Identify Required MCP Connector

Inspect MCP configuration files (read-only) to see what connectors exist. Report safe details only:

• Connector name
• Whether configuration appears present

NEVER display raw config values, passwords, API keys, tokens, or connection strings.

If the needed connector does not appear configured, flag it: "No matching MCP connector detected — this may block execution."

Label: Confirmed / Assumption / Unknown / Not confirmed

Step 6: Check for Existing Skill or SOP

Delegate this check to existing-asset-finder (pass the restated request from Step 1 as its input) rather than re-scanning .claude/skills/ independently — it owns overlap discovery project-wide, per the precedent record-a-skill-custom already established. Use its REUSE/EXTEND/MERGE/CREATE/STOP recommendation to fill this field: REUSE or EXTEND means a matching skill exists (note it); CREATE means none exists; MERGE or STOP should be flagged to the employee directly rather than silently folded into this map.

Label: Confirmed / Assumption / Unknown

Step 7: Define Allowed Actions & Permissions

Check CLAUDE.md for documented rules first. Ask via AskUserQuestion: "What are you allowed to do — any restrictions on what Claude can modify?" with options "No restrictions beyond standard safety rules", "Read-only — no changes without my approval", "Specific folders/files are off-limits", and "Need to check with my manager first".

Label: Confirmed / Assumption / Unknown | Use "Not confirmed" rather than inventing a permission.

Step 8: Define Validation Method

Ask via AskUserQuestion: "How will you check the output is correct?" with options "I'll review it manually", "Compare against a specific reference or example", "An automated check (tests/script)", and "Not sure yet".

Label: Confirmed / Assumption / Unknown | Use "Not confirmed" rather than inventing a validation rule.

Step 9: Identify Reviewer & Completion Evidence

Ask via AskUserQuestion, as two questions in one call: "Who needs to review the result?" (options: "Just me", "My manager or lead", "The original requester", "A specific team") and "What counts as done?" (options: "Reviewer approval", "Meets the stated validation criteria", "Delivered to the stated location", "Not sure yet").

Label: Confirmed / Assumption / Unknown / Not confirmed — use "Not confirmed" rather than inventing a reviewer.

Step 10: Identify Edge Cases & Alternate Paths

Ask via AskUserQuestion: "Are there edge cases or alternate paths this needs to handle — partial/missing data, error states, more than one valid way to do this?" with options "No unusual cases — straightforward", "Yes — I'll describe them", and "Not sure yet".

Label: Confirmed / Assumption / Unknown

Step 11: Readiness Decision

Based on all labeled fields, determine readiness:

Decision	Meaning
READY	All critical fields confirmed
READY WITH ASSUMPTIONS	Some fields are assumptions — proceed with caution
BLOCKED	Critical fields unknown — cannot proceed
NEEDS APPROVAL	Permissions or reviewer not confirmed
NEEDS CONNECTOR	Required MCP connector not found
NEEDS SOURCE CONFIRMATION	Data source not confirmed

Add one recommended next action based on the decision.

Output Format

After gathering all information, present the execution map in chat using this structure:

## Execution Map: [task name]

**Request:** [restated request] — [Confirmed]

### Business Goal
[goal] — [Confirmed/Assumption/Unknown]

### Expected Result
[result and format] — [Confirmed/Assumption/Unknown]

### Source of Truth
[data source] — [Confirmed/Assumption/Unknown]

### Required MCP Connector
[connector] — [Confirmed/Assumption/Unknown/Not confirmed]

### Existing Skill or SOP
[skill name or "None found"] — [Confirmed/Assumption/Unknown]

### Allowed Actions & Permissions
[permissions] — [Confirmed/Assumption/Unknown/Not confirmed]

### Validation Method
[criteria] — [Confirmed/Assumption/Unknown/Not confirmed]

### Reviewer & Completion Evidence
[reviewer and evidence] — [Confirmed/Assumption/Unknown/Not confirmed]

### Edge Cases & Alternate Paths
[cases, or "None identified"] — [Confirmed/Assumption/Unknown]

### Readiness Decision: **[DECISION]**

**Recommended next action:** [one concrete step]

After showing the map in chat, ask via AskUserQuestion: "Does this look correct? I can save it to output/first-task-mapper/ for your records." — options Yes, save it (Recommended) / No, let me fix something first. If they want changes, clarify as free text what to fix, update the map, and ask again.

Write only after the user explicitly confirms. When saving, use the file path:

output/first-task-mapper/[task-name]-execution-map.md

Write only inside output/first-task-mapper/ — never elsewhere.

Notes

• NEVER execute the mapped task — this skill only creates the plan
• NEVER modify project files, .claude/ configuration, MCP settings, or any project source
• NEVER display raw config values, passwords, API keys, tokens, or connection strings
• NEVER assume a reviewer, source, permission, or validation rule — use "Not confirmed" instead
• NEVER write outside output/first-task-mapper/
• Always show the execution map in chat first; save to file only after user confirms
• If the request is too vague, ask focused clarifying questions — never guess
• Inspect CLAUDE.md and existing skills before suggesting new workflows
• If project documentation is missing, flag it as a limitation and continue with questions
• Clickable-question convention: Steps 1-10 already use AskUserQuestion with drafted plausible options plus a free-text "Other". Converted the remaining plain-text save confirmation. Only the initial request description (when $ARGUMENTS is absent) and any "Other"/fix-up follow-ups stay free text.