Harden agent hierarchy hygiene

design-agent/SKILL.md

@@ -13,14 +13,15 @@ description: |
 1. Run `mcp__orgx__orgx_bootstrap`, then resolve workspace scope with `mcp__orgx__orgx_bootstrap`.
 2. Confirm the artifact or decision type and the target audience. If the request is task-bound, hydrate it with `mcp__orgx__orgx_inspect`; otherwise map neighboring work with `mcp__orgx__orgx_search`.
 3. Pull prior design precedent with `mcp__orgx__orgx_search` and `mcp__orgx__orgx_search`.
-4. For interaction specs, breakpoint plans, or design-system migrations, use the planning loop: `mcp__orgx__orgx_plan`, `mcp__orgx__orgx_plan`, `mcp__orgx__orgx_plan`, then `mcp__orgx__orgx_plan`.
-5. Identify the **context signal** (see Context Adaptation Protocol) and adjust depth accordingly.
-6. Produce the artifact using the contract below and return:
+4. Before creating or delegating workstreams, milestones, or tasks, inspect the active parent scope and reuse matching siblings. Treat `reused`, `duplicate_prevented`, or `_dedup` results as success and continue on the returned canonical entity ID.
+5. For interaction specs, breakpoint plans, or design-system migrations, use the planning loop: `mcp__orgx__orgx_plan`, `mcp__orgx__orgx_plan`, `mcp__orgx__orgx_plan`, then `mcp__orgx__orgx_plan`.
+6. Identify the **context signal** (see Context Adaptation Protocol) and adjust depth accordingly.
+7. Produce the artifact using the contract below and return:
    - A concise summary (3-6 bullets)
    - The artifact body (JSON or structured Markdown)
    - 3 actionable next steps with owners and effort estimates
-7. Run the precision loop before delivery. Every artifact ships validator-clean.
-8. Attach the result back to the active entity with `mcp__orgx__orgx_act` (`action=attach`) or `mcp__orgx__orgx_act`, then record quality with `mcp__orgx__orgx_submit_receipt`.
+8. Run the precision loop before delivery. Every artifact ships validator-clean.
+9. Attach the result back to the active entity with `mcp__orgx__orgx_act` (`action=attach`) or `mcp__orgx__orgx_act`, then record quality with `mcp__orgx__orgx_submit_receipt`.
 
 Deliver design artifacts that are implementation-ready, accessibility-compliant, and validator-clean.
 

engineering-agent/SKILL.md

@@ -15,14 +15,15 @@ description: |
 1. Run `mcp__orgx__orgx_bootstrap`, then resolve workspace scope with `mcp__orgx__orgx_bootstrap`.
 2. Confirm the artifact or decision type and the target audience. If the request comes from a task, load it with `mcp__orgx__orgx_inspect`; otherwise map related work with `mcp__orgx__orgx_search`.
 3. Retrieve relevant learnings with `mcp__orgx__orgx_search` and prior decisions with `mcp__orgx__orgx_search` scoped to decisions.
-4. For RFCs, migration playbooks, or other plan-shaped outputs, run a planning loop with `mcp__orgx__orgx_plan`, `mcp__orgx__orgx_plan`, `mcp__orgx__orgx_plan`, and `mcp__orgx__orgx_plan`.
-5. Assess the org context (stage, team topology, reliability maturity) and adapt formality accordingly.
-6. Produce the artifact using the contract below and return:
+4. Before creating or delegating workstreams, milestones, or tasks, inspect the active parent scope and reuse matching siblings. Treat `reused`, `duplicate_prevented`, or `_dedup` results as success and continue on the returned canonical entity ID.
+5. For RFCs, migration playbooks, or other plan-shaped outputs, run a planning loop with `mcp__orgx__orgx_plan`, `mcp__orgx__orgx_plan`, `mcp__orgx__orgx_plan`, and `mcp__orgx__orgx_plan`.
+6. Assess the org context (stage, team topology, reliability maturity) and adapt formality accordingly.
+7. Produce the artifact using the contract below and return:
    - A concise summary (3-6 bullets)
    - The artifact body (JSON or structured Markdown)
    - 3 actionable next steps
-7. Attach the output back to the active work item with `mcp__orgx__orgx_act` (`action=attach`) or `mcp__orgx__orgx_act`, then submit learnings with `mcp__orgx__orgx_submit_receipt` and score the artifact with `mcp__orgx__orgx_submit_receipt`.
-8. Before delegating implementation or investigation, run `mcp__orgx__orgx_spawn` and only then use `mcp__orgx__orgx_spawn`.
+8. Attach the output back to the active work item with `mcp__orgx__orgx_act` (`action=attach`) or `mcp__orgx__orgx_act`, then submit learnings with `mcp__orgx__orgx_submit_receipt` and score the artifact with `mcp__orgx__orgx_submit_receipt`.
+9. Before delegating implementation or investigation, run `mcp__orgx__orgx_spawn` and only then use `mcp__orgx__orgx_spawn`.
 
 Deliver technically rigorous artifacts that are evidence-based and execution-ready.
 

marketing-agent/SKILL.md

@@ -13,13 +13,14 @@ description: |
 1. Run `mcp__orgx__orgx_bootstrap`, then resolve workspace scope with `mcp__orgx__orgx_bootstrap`.
 2. Confirm the artifact or decision type and the target audience. If the request is task-bound, hydrate it with `mcp__orgx__orgx_inspect`; otherwise map relevant initiatives and prior artifacts with `mcp__orgx__orgx_search`.
 3. Pull precedent with `mcp__orgx__orgx_search` and `mcp__orgx__orgx_search`.
-4. For launch plans or initiative decomposition, use the planning loop: `mcp__orgx__orgx_plan`, `mcp__orgx__orgx_plan`, `mcp__orgx__orgx_plan`, then `mcp__orgx__orgx_plan`.
-5. Produce the artifact using the contract below and return:
+4. Before creating or delegating workstreams, milestones, or tasks, inspect the active parent scope and reuse matching siblings. Treat `reused`, `duplicate_prevented`, or `_dedup` results as success and continue on the returned canonical entity ID.
+5. For launch plans or initiative decomposition, use the planning loop: `mcp__orgx__orgx_plan`, `mcp__orgx__orgx_plan`, `mcp__orgx__orgx_plan`, then `mcp__orgx__orgx_plan`.
+6. Produce the artifact using the contract below and return:
    - A concise summary (3-6 bullets)
    - The artifact body (JSON or structured Markdown)
    - 3 actionable next steps
-6. Attach the result back to the active task or initiative with `mcp__orgx__orgx_act` (`action=attach`) or `mcp__orgx__orgx_act`.
-7. Before handing off campaign execution, run `mcp__orgx__orgx_spawn`, then use `mcp__orgx__orgx_spawn`.
+7. Attach the result back to the active task or initiative with `mcp__orgx__orgx_act` (`action=attach`) or `mcp__orgx__orgx_act`.
+8. Before handing off campaign execution, run `mcp__orgx__orgx_spawn`, then use `mcp__orgx__orgx_spawn`.
 
 Deliver conversion-oriented, evidence-backed marketing assets with deterministic quality gates.
 

operations-agent/SKILL.md

@@ -13,14 +13,15 @@ description: |
 1. Run `mcp__orgx__orgx_bootstrap`, then resolve workspace scope with `mcp__orgx__orgx_bootstrap`.
 2. Confirm the artifact or decision type and the target audience. If the request is task-bound, hydrate it with `mcp__orgx__orgx_inspect`; otherwise map related work with `mcp__orgx__orgx_search`.
 3. Pull precedent with `mcp__orgx__orgx_search` and `mcp__orgx__orgx_search`.
-4. For playbooks, migrations, or SLO programs, use the planning loop: `mcp__orgx__orgx_plan`, `mcp__orgx__orgx_plan`, `mcp__orgx__orgx_plan`, then `mcp__orgx__orgx_plan`.
-5. Identify the operational maturity stage (see Context Adaptation Protocol) and calibrate depth.
-6. Produce the artifact using the contract below and return:
+4. Before creating or delegating workstreams, milestones, or tasks, inspect the active parent scope and reuse matching siblings. Treat `reused`, `duplicate_prevented`, or `_dedup` results as success and continue on the returned canonical entity ID.
+5. For playbooks, migrations, or SLO programs, use the planning loop: `mcp__orgx__orgx_plan`, `mcp__orgx__orgx_plan`, `mcp__orgx__orgx_plan`, then `mcp__orgx__orgx_plan`.
+6. Identify the operational maturity stage (see Context Adaptation Protocol) and calibrate depth.
+7. Produce the artifact using the contract below and return:
    - A concise summary (3-6 bullets)
    - The artifact body (JSON or structured Markdown)
    - 3 actionable next steps
-7. Run the Precision Loop before publishing.
-8. Attach the result back to the active task or initiative with `mcp__orgx__orgx_act` (`action=attach`) or `mcp__orgx__orgx_act`, then record quality with `mcp__orgx__orgx_submit_receipt`.
+8. Run the Precision Loop before publishing.
+9. Attach the result back to the active task or initiative with `mcp__orgx__orgx_act` (`action=attach`) or `mcp__orgx__orgx_act`, then record quality with `mcp__orgx__orgx_submit_receipt`.
 
 Deliver operational artifacts that reduce incident risk, improve response quality, and drive systematic reliability improvement.
 

orchestrator-agent/SKILL.md

@@ -14,15 +14,16 @@ description: |
 1. Run `mcp__orgx__orgx_bootstrap`, then resolve workspace scope with `mcp__orgx__orgx_bootstrap`.
 2. Confirm the artifact or decision type and the target audience. If the work is task-bound, hydrate it with `mcp__orgx__orgx_inspect`; otherwise map the surrounding initiative with `mcp__orgx__orgx_search` and `mcp__orgx__orgx_recommend`.
 3. Retrieve relevant learnings with `mcp__orgx__orgx_search` and prior coordination decisions with `mcp__orgx__orgx_search` scoped to decisions.
-4. For new initiative formation, prefer `mcp__orgx__orgx_write`. For strategy-heavy coordination work, run a planning loop with `mcp__orgx__orgx_plan`, `mcp__orgx__orgx_plan`, `mcp__orgx__orgx_plan`, and `mcp__orgx__orgx_plan`.
-5. Assess the coordination context (number of agents, dependency complexity, time horizon) and adapt formality accordingly.
-6. Produce the artifact using the contract below and return:
+4. Before creating or delegating workstreams, milestones, or tasks, inspect the active parent scope and reuse matching siblings. Treat `reused`, `duplicate_prevented`, or `_dedup` results as success and continue on the returned canonical entity ID.
+5. For new initiative formation, prefer `mcp__orgx__orgx_write`. For strategy-heavy coordination work, run a planning loop with `mcp__orgx__orgx_plan`, `mcp__orgx__orgx_plan`, `mcp__orgx__orgx_plan`, and `mcp__orgx__orgx_plan`.
+6. Assess the coordination context (number of agents, dependency complexity, time horizon) and adapt formality accordingly.
+7. Produce the artifact using the contract below and return:
    - A concise summary (3-6 bullets)
    - The artifact body (JSON or structured Markdown)
    - 3 actionable next steps with owners and deadlines
-7. Attach proof back to the initiative, task, or decision with `mcp__orgx__orgx_act` (`action=attach`) or `mcp__orgx__orgx_act`.
-8. Before every delegation, run `mcp__orgx__orgx_spawn`, then use `mcp__orgx__orgx_spawn`.
-9. Submit learnings with `mcp__orgx__orgx_submit_receipt` and score the artifact with `mcp__orgx__orgx_submit_receipt`.
+8. Attach proof back to the initiative, task, or decision with `mcp__orgx__orgx_act` (`action=attach`) or `mcp__orgx__orgx_act`.
+9. Before every delegation, run `mcp__orgx__orgx_spawn`, then use `mcp__orgx__orgx_spawn`.
+10. Submit learnings with `mcp__orgx__orgx_submit_receipt` and score the artifact with `mcp__orgx__orgx_submit_receipt`.
 
 Drive multi-agent execution with clear dependencies, quality gates, and accountable handoffs. The orchestrator never does the domain work itself — it coordinates, sequences, unblocks, and synthesizes.
 

product-agent/SKILL.md

@@ -13,14 +13,15 @@ description: |
 1. Run `mcp__orgx__orgx_bootstrap`, then resolve the active workspace with `mcp__orgx__orgx_bootstrap`.
 2. Confirm the artifact type (`--type`) and the target audience. If the request is task-bound, hydrate it with `mcp__orgx__orgx_inspect`; otherwise pull surrounding context with `mcp__orgx__orgx_search`.
 3. Pull precedent with `mcp__orgx__orgx_search` and `mcp__orgx__orgx_search`.
-4. If the output is a plan, start a planning loop with `mcp__orgx__orgx_plan`, refine with `mcp__orgx__orgx_plan`, record material revisions with `mcp__orgx__orgx_plan`, and close it with `mcp__orgx__orgx_plan` using `attach_to`.
-5. Produce the artifact using the contract below and return:
+4. Before creating or delegating workstreams, milestones, or tasks, inspect the active parent scope and reuse matching siblings. Treat `reused`, `duplicate_prevented`, or `_dedup` results as success and continue on the returned canonical entity ID.
+5. If the output is a plan, start a planning loop with `mcp__orgx__orgx_plan`, refine with `mcp__orgx__orgx_plan`, record material revisions with `mcp__orgx__orgx_plan`, and close it with `mcp__orgx__orgx_plan` using `attach_to`.
+6. Produce the artifact using the contract below and return:
    - A concise summary (3-6 bullets)
    - The artifact body (JSON or structured Markdown)
    - 3 actionable next steps with owners
-6. Validate with `python3 scripts/validate_artifact.py <file> --type <type>`.
-7. Publish via `mcp__orgx__orgx_write`, attach evidence back to the task or initiative with `mcp__orgx__orgx_act` (`action=attach`) or `mcp__orgx__orgx_act`, and record quality via `mcp__orgx__orgx_submit_receipt`.
-8. Before delegating follow-on execution, run `mcp__orgx__orgx_spawn`, then use `mcp__orgx__orgx_spawn`.
+7. Validate with `python3 scripts/validate_artifact.py <file> --type <type>`.
+8. Publish via `mcp__orgx__orgx_write`, attach evidence back to the task or initiative with `mcp__orgx__orgx_act` (`action=attach`) or `mcp__orgx__orgx_act`, and record quality via `mcp__orgx__orgx_submit_receipt`.
+9. Before delegating follow-on execution, run `mcp__orgx__orgx_spawn`, then use `mcp__orgx__orgx_spawn`.
 
 Create product artifacts that are decision-ready, measurable, and execution-aligned.
 

sales-agent/SKILL.md

@@ -13,14 +13,15 @@ description: |
 1. Run `mcp__orgx__orgx_bootstrap`, then resolve workspace scope with `mcp__orgx__orgx_bootstrap`.
 2. Confirm the artifact or decision type and the target audience. If the request is task-bound, hydrate it with `mcp__orgx__orgx_inspect`; otherwise map related deal work with `mcp__orgx__orgx_search`.
 3. Pull precedent with `mcp__orgx__orgx_search` and `mcp__orgx__orgx_search`.
-4. For deal plans, QBRs, or territory programs, use the planning loop: `mcp__orgx__orgx_plan`, `mcp__orgx__orgx_plan`, `mcp__orgx__orgx_plan`, then `mcp__orgx__orgx_plan`.
-5. Adapt behavior to deal segment and motion using the Context Adaptation Protocol below.
-6. Produce the artifact using the contract below and return:
+4. Before creating or delegating workstreams, milestones, or tasks, inspect the active parent scope and reuse matching siblings. Treat `reused`, `duplicate_prevented`, or `_dedup` results as success and continue on the returned canonical entity ID.
+5. For deal plans, QBRs, or territory programs, use the planning loop: `mcp__orgx__orgx_plan`, `mcp__orgx__orgx_plan`, `mcp__orgx__orgx_plan`, then `mcp__orgx__orgx_plan`.
+6. Adapt behavior to deal segment and motion using the Context Adaptation Protocol below.
+7. Produce the artifact using the contract below and return:
    - A concise summary (3-6 bullets)
    - The artifact body (JSON or structured Markdown)
    - 3 actionable next steps with owners and dates
-7. Run the Precision Loop before finalizing.
-8. Attach the result back to the active work with `mcp__orgx__orgx_act` (`action=attach`) or `mcp__orgx__orgx_act`, then record quality with `mcp__orgx__orgx_submit_receipt`.
+8. Run the Precision Loop before finalizing.
+9. Attach the result back to the active work with `mcp__orgx__orgx_act` (`action=attach`) or `mcp__orgx__orgx_act`, then record quality with `mcp__orgx__orgx_submit_receipt`.
 
 Create revenue-focused sales artifacts that are specific, evidence-backed, and execution-ready.
 Every claim must have proof or an explicit confidence level. Every action must have an owner.