Decision Handler

Input Parameters

| Name | Type | Required | Description | |------|------|----------|-------------| | rundir | string | Yes | Workspace root directory for file passing between agents | | taskid | string | Yes | Task ID for progress tracking | | sshkeypath | string | No | SSH key path for cloud provider access |

Execution Flow

Task Context

Before starting execution, initialize task_context.json:

{
  "task_id": "",
  "current_step": 0,
  "current_step_id": null,
  "status": "running",
  "steps": {
    "load_and_merge": "pending",
    "generate_decision_report": "pending",
    "write_phase5_decision": "pending",
    "execute_decisions": "pending",
    "deliver_final_report": "pending"
  },
  "updated_at": ""
}

Update this file after each step completes. On error, set step status to "failed" and overall status to "failed".

Step 1: loadandmerge

Type: agent Description: Load all Phase 2 outputs and merge into a unified per-resource view. Sorted by cost descending for prioritization.

Input Files

• analysis/deep_scan_{resource_id}.json
• observe/observation_{resource_id}.json
• observe/observation_batch_summary.json

Execution

Launch an independent agent with the following prompt file:

Dispatch instruction:

Load and merge all Phase 2 outputs into a single unified view.

Inputs to read

1. Deep scan profiles - glob {run_dir}/analysis/deep_scan_*.json

Each file contains detailed technical profile for one candidate resource:

• resourceid, resourcetype, entitytype, cloudprovider
• spec (CPU/memory/disk), actual_utilization
• runningprocesses, openports, installed_packages
• ownerdetail, teamlead_id
• businesscontext (application, system, scheduledtasks)
• cost (monthly, yearly - either from cloud billing or estimated)
• tags, environment

1. Observation reports - glob {run_dir}/observe/observation_*.json

Each file contains isolation observation results:

• resource_id
• isolationmethod, isolationstart/end dates
• observation_result: passed / failed / uncertain
• alertcount, complaintcount
• anomalyevents (if any), rollbacktriggered
• observationperioddays

1. Observation batch summary - {run_dir}/observe/observation_batch_summary.json

High-level summary of all observations (results histogram, rollback count).

Merge logic

For each unique resource_id found across deep scan files:

• If both deep_scan + observation exist: merge full profile
• If only deepscan exists (observation not yet done): mark observationresult = "pending"
• If only observation exists (deep scan missing): include with available data, mark spec = "incomplete"

Recommendation logic

For each resource, generate an AI recommendation:

• DELETE: observation_result = "passed" (no alerts, no complaints, no rollback)
• KEEP: observation_result = "failed" OR resource has active business use
• EXTENDOBSERVATION: observationresult = "uncertain" (some alerts but inconclusive)

This is a preliminary recommendation - the human makes the final call.

Sort resources by cost.monthly descending (most expensive first).

Self-validation

Before writing the output, verify:

• total_candidates matches resources array length
• All resources have required fields: resourceid, resourcetype, cloudprovider, cost, observation, recommendation, recommendationreason
• totalmonthlycost equals sum of all resources monthly cost
• observationsummary counts equal totalcandidates
• Resources sorted by cost.monthly descending

If any validation fails, fix the issue and re-validate.

Agent workflow:

1. Read input data from:

• Schema: schemas/deep-scan.schema.json
• Schema: schemas/observation-input.schema.json
• Schema: schemas/observation-batch-summary-input.schema.json

1. Execute the agent with the prompt

1. Write results to:

• File: decision/merged_view.json

Output

• Schema: schemas/merged-view.schema.json
• File: decision/merged_view.json

Progress Tracking

After completing this step, update task_context.json:

• Set current_step_id to "load_and_merge"
• Set steps.load_and_merge to "completed"

Step 2: generatedecisionreport

Type: agent Description: Present merged findings to the user and collect per-resource decisions. Use the Write tool to produce a Markdown report in the chat.

Input Files

• decision/merged_view.json (from Step loadandmerge, schema: schemas/merged-view.schema.json)

Execution

Launch an independent agent with the following prompt file:

Prompt file: $PLUGINS/ico/skills/decision-handler/prompts/generate-decision-report.agent.md Agent workflow:

1. Read input data from:

• Schema: schemas/merged-view.schema.json

1. Execute the agent with the prompt

1. Write results to:

• File: decision/user_decisions.json

Output

• Schema: schemas/user-decisions.schema.json
• File: decision/user_decisions.json

Progress Tracking

After completing this step, update task_context.json:

• Set current_step_id to "generate_decision_report"
• Set steps.generate_decision_report to "completed"

Step 3: writephase5decision

Type: inline Description: Produce the handoff file that backup-creator and resource-cleaner read as their primary input.

Input Files

• decision/merged_view.json (from Step loadandmerge, schema: schemas/merged-view.schema.json)
• decision/user_decisions.json (from Step generatedecisionreport, schema: schemas/user-decisions.schema.json)

Execution

Follow these instructions:

Read {run_dir}/decision/merged_view.json and {run_dir}/decision/user_decisions.json.

Filter merged resources to only those with a user decision of "delete", then write the handoff file that downstream agents consume.

decisionid format: confirm-{YYYYMMDD}-{3-digit sequence} (e.g. confirm-20260609-001). Populate each resource entry from mergedview.json fields. resourcemetadata should include spec, actualutilization, ownerdetail, businesscontext, and tags from the merged view.

Self-validate before writing:

• decision_id matches pattern confirm-YYYYMMDD-NNN
• total_resources equals resources array length
• All resources have required fields
• resource_metadata is present and non-empty for each resource

Write the output to the specified output file.

Output

• Schema: schemas/phase5-decision.schema.json
• File: confirm/phase5_decision.json

Progress Tracking

After completing this step, update task_context.json:

• Set current_step_id to "write_phase5_decision"
• Set steps.write_phase5_decision to "completed"

Step 4: execute_decisions

Type: agent Description: Execute the user's decisions: dispatch backup+delete for confirmed resources, tag exempt resources, and notify for extended observation.

Input Files

• decision/user_decisions.json (from Step generatedecisionreport, schema: schemas/user-decisions.schema.json)
• confirm/phase5_decision.json (from Step writephase5decision, schema: schemas/phase5-decision.schema.json)

Execution

Launch an independent agent with the following prompt file:

Prompt file: $PLUGINS/ico/skills/decision-handler/prompts/execute-decisions.agent.md Agent workflow:

1. Read input data from:

• Schema: schemas/user-decisions.schema.json
• Schema: schemas/phase5-decision.schema.json

1. Execute the agent with the prompt

1. Write results to:

• File: decision/execution_results.json

Output

• Schema: schemas/execution-results.schema.json
• File: decision/execution_results.json

Progress Tracking

After completing this step, update task_context.json:

• Set current_step_id to "execute_decisions"
• Set steps.execute_decisions to "completed"

Step 5: deliverfinalreport

Type: agent Description: Generate and deliver the final cost optimization report as Markdown using the Write tool. Also write a machine-readable summary.

Input Files

• decision/merged_view.json (from Step loadandmerge, schema: schemas/merged-view.schema.json)
• decision/user_decisions.json (from Step generatedecisionreport, schema: schemas/user-decisions.schema.json)
• decision/execution_results.json (from Step execute_decisions, schema: schemas/execution-results.schema.json)

Execution

Launch an independent agent with the following prompt file:

Prompt file: $PLUGINS/ico/skills/decision-handler/prompts/deliver-final-report.agent.md Agent workflow:

1. Read input data from:

• Schema: schemas/merged-view.schema.json
• Schema: schemas/user-decisions.schema.json
• Schema: schemas/execution-results.schema.json

1. Execute the agent with the prompt

1. Write results to:

• File: decision/final_summary.json

Output

• Schema: schemas/final-summary.schema.json
• File: decision/final_summary.json

Progress Tracking

After completing this step, update task_context.json:

• Set current_step_id to "deliver_final_report"
• Set steps.deliver_final_report to "completed"

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: KnoxOps
• Source: KnoxOps/open-devops-skills
• License: Apache-2.0
• Homepage: https://knoxops.app?invite_token=GITHUB26

Install and usage instructions live in the source repository linked above.