imabusyman/autonomous-agent-harness
prompts/squad/skills/autonomous-agent-harness/SKILL.md
Provide a task queue, scheduler, and persistent memory system for autonomous multi-agent loops. Track task state, monitor agent progress, retry on failure, and persist decisions across sessions. Use this in the Orchestrator parallel path when running an autonomous PRD-driven loop without human gates.
testing
Updated Apr 25, 2026
$ install --global
skillsauthnpx skillsauth add imabusyman/CodebrewRouter autonomous-agent-harnessInstall this skill globally with one command. Works with Claude Code, Cursor, and Windsurf.
Security Scan Results
3 of 9 scanners reported clean
Some scanners were skipped, did not run, or reported a non-clean status. Review each row below.
3
Scanners Passed
9
Scanners in report
Clean
TrivyContainer and dependency vulnerability scanner
95%
Clean
SemgrepStatic code analysis for vulnerabilities
95%
Clean
mcp-scan (Snyk)Model Context Protocol security validation
95%
Skipped
Snyk (dep)Open source security scanning
50%
Skipped
Socket.devSupply chain security analysis
50%
Skipped
VirusTotalMulti-engine malware detection
50%
Skipped
CrowdStrikeAdvanced threat intelligence
50%
Skipped
OSV-ScannerOpen Source Vulnerability database check
50%
Skipped
OWASP Dep-Check
50%
Last scanned: Apr 25, 2026, 7:19 PM141.7s1 file scanned
SKILL.md
- name:
- autonomous-agent-harness
- description:
- Provide a task queue, scheduler, and persistent memory system for autonomous multi-agent loops. Track task state, monitor agent progress, retry on failure, and persist decisions across sessions. Use this in the Orchestrator parallel path when running an autonomous PRD-driven loop without human gates.
Autonomous Agent Harness — task queue + scheduler + memory
A runtime framework for fully autonomous agent workflows (no human gates). Manages task queues, agent scheduling, persistent decision logs, and failure recovery.
When to use
- Parallel path: Orchestrator is running autonomously against a PRD (no human
/agent squadgate). - You need task queue management (prioritize, retry, skip on failure).
- Agent decisions must persist across sessions (in case of restart/recovery).
- You want monitoring dashboards (task status, agent health, completion %).
Components
1. Task queue
Initial state (from PRD):
Queue:
[pending] Task 1: Implement Provider X (assigned: Coder)
[pending] Task 2: Unit test Provider X (assigned: Tester)
[pending] Task 3: Wire Provider X in AppHost (assigned: Infra)
State transitions:
pending → assigned → in_progress → done / blocked / failed → (retry or skip)
2. Scheduler
- Pull tasks from queue in priority order (dependencies first).
- Assign agents (Coder for impl, Tester for tests, Infra for config).
- Dispatch handoffs (emit structured envelope; agent works independently).
- Monitor for completion (poll for
[DONE]or[BLOCKED]). - Advance or backtrack (move to next task or fail-retry).
3. Persistent memory (reasoning log)
Append-only decision log under Docs/squad/runs/<ts>/:
## 2026-04-20T15:45:00Z — scheduler — MEDIUM
Decision: Task 2 (tests) was skipped due to missing implementation.
Rationale: Quality gate requires tests to run after impl completes.
Evidence: Task 1 still in [in_progress]; Task 2 moved to [queued].
Recovery across sessions: On restart, read the log, resume from the last checkpoint.
4. Failure recovery
- Failed task: Move to
[blocked]+ emit alert. Orchestrator decides: retry, skip, or abort. - Failed quality gate: Stay in
[in_progress], loop back to agent with fixes needed. - Persist state: If the session dies, recovery reads the log and resumes exactly where it was.
Orchestrator integration
The Orchestrator (Conductor's parallel-path sibling) drives the harness:
- Load PRD → generate task queue.
- Loop:
- Pull next ready task from queue.
- Dispatch to appropriate agent (Coder / Tester / Infra).
- Await
[DONE]or[BLOCKED]. - Update queue + reasoning log.
- Retry or move to next task.
- When queue is empty + all tasks
[done], trigger Reviewer + Security-Review.
Configuration
harness:
max_retries: 3
task_timeout: 1h # per task
session_timeout: 8h # total runtime before auto-checkpoint
priority_rules:
- "Architect tasks (ADRs) run first"
- "Coder tasks (impl) run before Tester"
- "Infra tasks (config) run last"
failure_mode: "retry_once_then_block" # or "fail_fast"
See also
loop-operator— supervise the autonomous loop, detect stalls, safe interventions.subagent-driven-development— structure subagent delegation.dispatching-parallel-agents— compute task sets from plan.
Related Skills
imabusyman/write-a-prd
development
VerifiedTrustedCommunity
Generate a Product Requirements Document from a task description, user intent, or PRD outline. Structure the PRD with sections: Overview, Problem Statement, Goals, Scope, Features, Acceptance Criteria, Metrics. Use this when the Planner or Orchestrator needs to formalize requirements before decomposing into implementation steps.
SKILL.mdUpdated Apr 25, 2026
imabusyman/team-builder
development
VerifiedTrustedCommunity
Compose parallel teams of specialized agents for a multi-faceted task. Map subtasks to agent roles, balance workload, and validate team composition. Use this when breaking a task into parallel streams that require different expertise (Coder, Tester, Infra, etc.).
SKILL.mdUpdated Apr 25, 2026
imabusyman/subagent-driven-development
development
VerifiedTrustedCommunity
Delegate implementation work to specialized subagent instances, each inheriting a task scope and constraints from the parent. Emit structured handoffs, monitor completion signals, aggregate results. Use this in the Orchestrator parallel path when spawning independent Coder/Tester/Infra subagents from a single coordinator.
SKILL.mdUpdated Apr 25, 2026
imabusyman/reasoning
testing
VerifiedTrustedCommunity
Maintain the append-only reasoning log and write handoff envelopes for every squad delegation. Use this when the Conductor delegates to a specialist or when a specialist records a non-trivial decision.
SKILL.mdUpdated Apr 25, 2026