Project Tracking
Operate on tasks and goals in Epismo projects — from a quick status update to full goal restructuring.
Core principle: read current state → apply the smallest useful change.
For connection setup, surface conventions, scope model, share URL resolution, and error handling, see Epismo Basics.
For workflow pack discovery and release, see Workflow Pack.
Surface selection: CLI and MCP connect to the same Epismo service. Use CLI if available; fall back to MCP if not. Never use both in the same session.
Operations
| Operation | CLI | MCP |
|---|
| epismo track search --type task|goal --filter '{...}'
| |
| epismo track get --id <id>
| |
| epismo track create --input '<json>'
| |
| epismo track update --id <id> --input '<json>'
| |
| epismo track delete --id <id>
| |
| epismo track apply --input '<json>'
| |
For filter keys, status values, entity relationships, and search recipes, see Search & Filter.
Intent Router
| Intent | When to choose | Steps | Key reference |
|---|
| Update existing work | Field change on something that already exists (status, assignee, date, priority, content) | 3 Coordinate — Partial Update | Runbook |
| Add one new item | Exactly one task or goal; existing structure is valid | 3 Coordinate — New Item | Runbook |
| Plan multiple items or a new goal | New goal structure, multi-step plan, or batch creation | 2 Plan → 3 Coordinate — Large-Scale | Runbook |
| Unblock or rebalance | Stalled queue, overloaded assignee, dependency bottleneck, noisy backlog | 4 Risk → 3 Coordinate — Recovery | Runbook |
| Design an AI-owned task | Delegating work to an AI agent with clear acceptance criteria | 3 Coordinate + AI Delegation | AI Delegation |
| Unclear intent | Cannot confidently match any row above | 1 Intake — ask once | — |
4-Step Flow
1 Intake
Confirm the user's desired outcome, constraints, timeline, active workspace, and target project scope.
Ask one clarifying question if intent is ambiguous. If unanswered, default to read-only and report findings.
2 Plan
Run before structural changes (new goals, multi-item plans, dependency rewiring).
Follow Runbook — Planning Checks: diagnose current state → select mode → design ownership contracts → confirm write destination → set next checkpoint.
Always choose the smallest mode that satisfies the request.
3 Coordinate
Pick one execution mode from Runbook — Mode Selector:
- Partial Update — localized field edits on existing entities.
- New Item Creation — exactly one new task / goal.
- Large-Scale Planning — multiple coordinated items with ownership contracts.
- Recovery — address overload, stall, or noisy backlog.
When a task status changes to
, run a downstream dependents check: query
and report what is now unblocked.
Operational state lives in tracks. Use workflow packs only when the structure should be reused beyond the current execution context — see Workflow Pack.
Design AI-owned tasks with AI Delegation.
Use Track Operations for structured write output after the Runbook checks are complete.
4 Risk
Evaluate bottlenecks and dependency risk per Runbook.
Run backlog hygiene (stale, duplicate, low-signal) during recovery.
Write Safety
- Scope first — confirm the target project before any write. If unclear, ask once.
- No silent writes — if a clarification question is unanswered, continue read-only.
- Approval for risk — require explicit approval for large-scale plans, multi-entity recovery, and destructive changes.
Source of Truth
Repository:
https://github.com/epismoai/skills