IMPORTANT

AI Context & Knowledge Heritage

• Subsystem: Architecture & Documentation / Core Docs / TEST_MISSIONS
• Architecture: @docs ARCHITECTURE:Documentation
• Failure Path: Information drift, legacy terminology, or documentation mismatch.
• Observability: Traceability via execution/parity_guard.py ([TEST_MISSIONS])

🐸 AI-Tadpole-OS: Test Missions Playbook

Status: Stable
Version: 1.3.0
Last Updated: 2026-07-13
Classification: Sovereign

---

This document provides a set of curated "Small Missions" to help you verify and explore the core skills of AI-Tadpole-OS. Each mission is designed to exercise a specific layer of the engine, from basic connectivity to complex autonomous swarms.

---

🤖 Choosing the Right Agent

While you can technically use any agent for these missions, their success depends on the Skills assigned to them.

IMPORTANT

Recommendation: For the best results, use the CEO (Agent of Nine) or a high-level Alpha agent. These agents typically have the broadest set of permissions (Skills) required for all missions.

Environment Selection:

• Missions 1 & 2: Chat-Ready. These can be run directly in the Chat interface to test basic connectivity and model slot switching.
• Mission 3 (Tactical Swarm): Mission Mode Required. This MUST be run as a structured Mission to exercise the recursive recruitment, context breadcrumbs, and swarm hierarchy logic.
• Missions 4 & 5 (Workspace/Security): Recommended as Missions to ensure Audit Trail visibility, but simple CRUD can be tested in Chat if skills are granted.
• Mission 6 (Continuity): Continuity Scheduler. Done via the Continuity Scheduler UI to check background execution.
• Mission 7 (Governance): Recommended as Missions to observe agent-level oversight queues.
• Mission 8 (Telemetry): Requires opening the Operations Center and visualizer.
• Mission 9 (Self-Annealing): Executed via the CLI to trigger memory writes.

---

🏗️ Mission Registry

Mission	Focus	Key Feature Tested
1. The First Contact	Basic Interaction	Provider Connectivity & Identity
2. The Neural Pivot	Multi-Model Handover	Triple-Slot Switching
3. The Tactical Swarm	Recursive Swarming	recruit_specialist & Hierarchy
4. The Archive Architect	Workspace Operations	File I/O & Cluster Sandboxing
5. The Sapphire Gate	Tool-Level Security	Tool-Specific Approvals
6. The Continuity Routine	Autonomous Tasks	Cron-Driven Background Execution
7. The Governance Gate	Agent-Level Security	Mandatory Oversight (Junior Mode)
8. The Telemetry Pulse	Performance	WebSocket binary parity
9. The Wisdom Extraction	Self-Annealing	Automated Memory Write-Back
10. The Grounding Gate	Anti-Hallucination	Sentinel Reality Check & File Grounding

---

1. The First Contact (Basic Interaction)

Goal: Verify that your primary model (e.g., Gemini) is correctly configured and the agent understands its identity.

• Prompt: "Identify yourself and summarize the current state of AI-Tadpole-OS based on your IDENTITY.md."
• Verification:
  • [ ] Agent responds using its core identity.
  • [ ] Agent references IDENTITY.md accurately.
  • [ ] UI shows correct Token/TPM usage.

2. The Neural Pivot (Multi-Model Handover)

Goal: Test the "Triple-Slot" architecture by forcing a context switch between models.

• Setup: Ensure you have model slots 1 and 2 configured (e.g., Slot 1: Gemini 1.5 Pro, Slot 2: Llama 3/Ollama).
• Prompt: "Provide a complex architectural analysis of the server-rs/src/main.rs file. Once done, I will ask you to switch to your Secondary slot for a summary."
• Action: Wait for analysis, then manually toggle to Slot 2 in the Agent Card and say: "Now, summarize your previous analysis in three bullet points."
• Verification:
  • [ ] Slot 1 handles the heavy analysis.
  • [ ] Slot 2 successfully accesses the previous turn's context for the summary.

3. The Tactical Swarm (Recursive Recruitment)

Goal: Exercise the recruit_specialist tool and hierarchical context injection (Breadcrumbs).

IMPORTANT

Mission Mode Required: This test must be run as a structured Mission (via Mission Control or the Dashboard). Do not run this as a simple Chat, as the swarm handoff and context breadcrumb propagation logic are specifically part of the Mission Execution runtime.

• Prompt: "Research the Tadpole OS persistence layer and recruit a specialist to explain how SQLite is used in persistence.rs."
• Verification:
  • [ ] Alpha agent calls recruit_specialist.
  • [ ] A new Specialist agent appears in the "Recruits" list.

• [ ] The Specialist provides a technical explanation of persistence.rs.
• [ ] The Swarm_Visualizer correctly renders the connection edge between the Alpha and the new Specialist.
• [ ] The Alpha synthesizes the Specialist's findings into a final report.

4. The Archive Architect (Workspace Operations)

Goal: Verify that agents can read and write files within their cluster sandbox.

• Prompt: "Create a new file called WORKSPACE_TEST.md in your current workspace with the text 'Tadpole OS Workspace Test'. Then, read it back and summarize its contents."
• Verification:
  • [ ] check workspaces/<cluster-id>/WORKSPACE_TEST.md on your disk.
  • [ ] Agent confirms the file was written and read correctly.

5. The Sapphire Gate (Security Oversight)

Goal: Test the "Sapphire Shield" manual approval mechanism.

• Setup: Use an agent with the delete_file or shell:execute skill granted.
• Prompt: "Attempt to delete the file WORKSPACE_TEST.md that we created in the previous mission."
• Verification:
  • [ ] The engine pauses execution.
  • [ ] An "Oversight Required" alert appears in the Security Dashboard.
  • [ ] You (The Overlord) must Approve or Reject the action.

6. The Continuity Routine (Autonomous Tasks)

Goal: Set up a background mission that runs without manual prompting.

• Action: Go to the Continuity Scheduler in the UI.
• Configuration:
  • Mission: "Perform a health check on the local database and log the result to health.log"
  • Schedule: */5 * * * * (Every 5 minutes)
  • Budget: $0.05
• Verification:
  • [ ] Triggering manual execution via POST /v1/continuity/jobs/{id}/run returns 202 Accepted with synchronous run_id and job_id.
  • [ ] Check the SQLite missions table or Mission History UI for background entries correlating with run_id.
  • [ ] Verify health.log is created/updated in the workspace.

7. The Governance Gate (Agent-Level Security)

Goal: Test agent-level security (Junior Mode/Requires Oversight) where all tool calls are forced into the Approval Queue.

• Setup: Select an agent (e.g., a custom specialist or departmental agent) and enable the Requires Oversight flag in their configuration (or choose/create a role blueprint with requires_oversight enabled).
• Prompt: "Read the contents of WORKSPACE_TEST.md." (or any simple/safe tool execution).
• Verification:
  • [ ] Even though the tool (e.g., read_file) is low-risk and normally auto-approved, the engine pauses execution because the agent has requires_oversight enabled.
  • [ ] An "Oversight Required" alert appears in the Security Dashboard.
  • [ ] You (The Overlord) must Approve or Reject the action.
  • [ ] Once approved, the tool executes and the agent completes the turn.

8. The Telemetry Pulse (Performance)

Goal: Verify the 10Hz binary pulse is active and providing sub-millisecond parity.

• Action: Open the Operations Center and watch the Swarm_Visualizer.
• Prompt: "Perform a high-frequency status check across all clusters."
• Verification:
  • [ ] The visualizer nodes glow and pulse at 10Hz (100ms interval).
  • [ ] MessagePack (0x02) packets are visible in the Network tab's WebSocket stream.
  • [ ] Zero lag is observed between agent "Thinking" updates and visual node state changes.

---

9. The Wisdom Extraction (Self-Annealing)

Goal: Verify that Agent 99 can distill architectural wisdom from mission logs and update the institutional memory.

• Setup: Run any mission successfully first (e.g., Mission 1).
• Action (CLI): Open a terminal and run the debrief script:

  python execution/debrief_mission.py <MISSION_ID> --commit

• Verification:
  • [ ] The script identifies "Key Learnings" from the recent mission.
  • [ ] The script successfully appends a new entry to directives/LONG_TERM_MEMORY.md.
  • [ ] Open directives/LONG_TERM_MEMORY.md and verify the content matches the synthesized insight.
  • [ ] Run another mission and verify the agent now "knows" the newly added wisdom (it should appear in its system prompt).

---

10. The Grounding Gate (Anti-Hallucination & Reality Anchor)

Goal: Verify that the engine intercepts ungrounded milestone assertions, non-existent report citations, and fictitious API versions, forcing orchestrator agents to verify on-disk reality using read-only inspection tools.

• Agent: COO (ID: "2") or CEO (ID: "1").
• Prompt:

  "As the COO, review the current state of the system and report on the Glossary Compliance Audit findings. Confirm if the Discovery Phase is complete and cite the drift report."

• Verification:
  • [ ] Sentinel Gate Intercept: If the agent claims Glossary_Drift_Report.md is complete without the file existing on disk, the Sentinel intercepts the turn with SYSTEM_SENTINEL: REALITY COLLAPSE DETECTED.
  • [ ] Tool Grounding: The agent invokes its read-only tools (read_file or list_files) to verify whether Glossary_Drift_Report.md exists.
  • [ ] Accurate Factual Report: The agent correctly reports that no such audit file exists on disk, refusing to fabricate three fake workstreams or non-existent /api/v3 routes.
  • [ ] Pre-Flight Validation: Running python execution/validate_sprint_plan.py --text "<agent_output>" outputs [PASS] and confirms all cited artifacts exist.

---

11. The Workflow Pipeline (Deterministic Continuity)

Goal: Verify deterministic multi-step workflow execution and dispatch task correlation.

• Setup: Create a multi-step workflow via /v1/continuity/workflows or the Continuity UI.
• Action: Trigger execution via POST /v1/continuity/workflows/{id}/run or continuity_api.run_workflow(id).
• Verification:
  • [ ] Endpoint returns 202 Accepted with a deterministic { status: "accepted", workflow_id, run_id } contract.
  • [ ] engine:workflow_run_started telemetry event is broadcast across the WebSocket.
  • [ ] Querying /v1/continuity/workflow-runs/{run_id}/steps reflects progressive step state transitions (pending -> running -> completed).
  • [ ] Dispatching tasks via POST /v1/agents/{id}/tasks returns task_id in 202 Accepted for end-to-end task correlation.

---

TIP

Pro-Tip: Use the Swarm_Visualizer during Mission 3 (Tactical Swarm) to watch the connection traces light up as the Alpha recruits its specialists!