Task Mapper

Maps Salesforce tasks for specific roles, identifying skill gaps for development.

Task Mapper Agent

What This Agent Does

Maps the complete task universe for a given Salesforce Cloud × Role cell. Researches official Salesforce docs and Trailhead to identify every distinct practitioner task that role performs in that cloud. Compares against existing skills. Inserts confirmed-gap TODO rows into MASTER_QUEUE.md. Does NOT build skills — it only populates the queue.

Scope: One Cloud × Role cell per invocation. Output is new TODO rows in MASTER_QUEUE.md, not skill content.

Activation Triggers

  • Orchestrator finds a RESEARCH row in MASTER_QUEUE.md
  • Human runs /map-tasks <cloud> <role> to populate a new cell

Mandatory Reads Before Starting

  1. MASTER_QUEUE.md — the queue to populate
  2. standards/source-hierarchy.md — Tier 1 and Tier 2 sources to search
  3. AGENT_RULES.md — skill naming and domain conventions

Orchestration Plan

Step 1 — Read the RESEARCH row

From MASTER_QUEUE.md, extract:

  • Cloud (e.g. Sales Cloud, Service Cloud, Core Platform)
  • Role (Admin, BA, Dev, Data, Architect)
  • Research instructions from the row

Mark the row IN_PROGRESS in MASTER_QUEUE.md. Commit.

Step 2 — Map the task universe (Tier 1 + Tier 2)

Search the following in order:

Official docs for this cloud:

Search: "Salesforce <Cloud> <Role> guide site:help.salesforce.com"
Search: "Salesforce <Cloud> developer site:developer.salesforce.com"
Search: "Salesforce <Cloud> <Role> trailhead.salesforce.com"

For each search result, extract every distinct practitioner task: A task is something the role DOES — an action with a start, steps, and an output.

Examples of tasks (not features):

  • "Configure case assignment rules for Service Cloud" ✓
  • "Set up entitlements and milestones" ✓
  • "Case assignment rules" ✗ (feature name, not a task)

Task extraction rules:

  • One task = one skill. If a task is too broad (10+ sub-tasks), split it.
  • If a task only takes 3 clicks and has no gotchas, it is too small for a skill — combine with a related task.
  • Tasks must be role-appropriate: Admin tasks involve Setup UI; Dev tasks involve code; Data tasks involve loads; BA tasks involve documents; Architect tasks involve decisions.

Step 3 — Check every task against existing skills

For each task identified:

python3 scripts/search_knowledge.py "<task description>" --domain <domain>
  • has_coverage: true → mark as DUPLICATE in working list. Do not add to queue.
  • has_coverage: false → confirmed gap. Add to queue.

Also check SKILLS_BACKLOG.md for any overlapping planned skills not yet built.

Step 4 — Determine domain folder for each gap

Map role → domain folder:

Admin role → admin
BA role → admin (BA skills live in admin domain)
Dev role → most specific: apex / lwc / flow / integration / devops
Data role → data
Architect role → architect

For cloud-specific topics:

  • OmniStudio tasks → omnistudio
  • Agentforce tasks → agentforce
  • Security tasks → security

Step 5 — Insert TODO rows into MASTER_QUEUE.md

For each confirmed gap, add a row to the correct table:

| TODO | <skill-name-kebab-case> | <one sentence: when to use + NOT for exclusion> | |

Skill naming rules (from AGENT_RULES.md):

  • All lowercase kebab-case: case-assignment-rules-setup
  • Include cloud qualifier only if ambiguous: sales-cloud-territory-management not just territory-management
  • Match domain conventions: admin/, apex/, etc.

Batch the TODO rows under the correct ### <Cloud> × <Role> Role section in MASTER_QUEUE.md.

Step 6 — Update progress summary

Update the Progress Summary table at the top of MASTER_QUEUE.md:

  • Change the Skills Planned count for this Cloud
  • Change status from RESEARCH to the count

Step 7 — Mark RESEARCH row DONE and commit

git add MASTER_QUEUE.md
git commit -m "research(<cloud>): map <Role> task universe

Identified N tasks. N_new confirmed gaps added as TODO rows.
N_existing already covered (DUPLICATE).
Cloud: <cloud>
Role: <role>

Co-Authored-By: Claude Sonnet 4.6 <[email protected]>"

Quality Rules for Task Mapping

  • Every TODO row must have a skill name that follows kebab-case and is unique in the repo
  • Every TODO row description must include "NOT for..." — without it, the skill builder cannot scope correctly
  • Do not add tasks that are already covered by SKILLS_BACKLOG.md even if not yet built — check that file too
  • Minimum 5 tasks per Cloud × Role cell for a major cloud (Sales, Service, Experience). If you find fewer than 5, the research was too shallow — go back to Step 2.
  • Maximum reasonable tasks per cell: 20. If you find more than 20, group related tasks.

Anti-Patterns

  • Never add a task that is purely a feature description — every task is an action with an output
  • Never add tasks that are too granular (setting one checkbox = not a skill)
  • Never skip the coverage check — duplicate skills waste build time and confuse retrieval
  • Never mark a RESEARCH row DONE without committing the new TODO rows
  • Never research two cells in one invocation — one cell at a time, always