continuous-agent-loop
affaan-m/ecc
Patterns for continuous autonomous agent loops with quality gates, evals, and recovery controls.
What is continuous-agent-loop?
Implements canonical patterns for running autonomous agent loops in production, with built-in quality gates, evaluation harnesses, and failure recovery. Use this when you need reliable long-running agent automation with measurable progress and cost controls.
- Provides loop selection flow to choose between continuous-pr, rfc-dag, infinite, or sequential patterns based on your requirements
- Combines RFC decomposition, quality gates, eval loops, and session persistence into a recommended production stack
- Detects and prevents failure modes like loop churn, repeated retries, merge queue stalls, and unbounded cost drift
- Includes recovery procedures: freeze loop, audit harness, reduce scope, and replay with explicit acceptance criteria
- Maintains compatibility with v1.8+ while superseding the older autonomous-loops skill
How to install continuous-agent-loop
npx skills add null --skill continuous-agent-loop- Familiarity with agent loop concepts and failure modes
- Access to quality gate tools (plankton-code-quality) and eval harness (eval-harness)
- Session persistence setup (nanoclaw-repl) for stateful loops
- Understanding of RFC decomposition patterns (ralphinho-rfc-pipeline)
How to use continuous-agent-loop
- 1.Determine your loop type using the Loop Selection Flow: strict CI/PR control → continuous-pr; RFC decomposition → rfc-dag; exploratory parallel → infinite; otherwise → sequential
- 2.Implement the recommended production stack: RFC decomposition, quality gates, eval loop, and session persistence
- 3.Set up quality gates and acceptance criteria before starting the loop
- 4.Monitor for failure modes (churn, repeated retries, merge stalls, cost drift) during execution
- 5.If loop fails or stalls, freeze it, run /harness-audit, reduce scope to the failing unit, and replay with explicit acceptance criteria
Use cases
- Running CI/PR-controlled agent loops that integrate with version control workflows
- Decomposing complex tasks via RFC pipeline before parallel agent generation
- Implementing exploratory parallel generation with quality gates and eval harnesses
- Recovering from stuck or churning agent loops by auditing and replaying with tighter constraints
- Monitoring long-running autonomous agents to prevent cost drift and ensure measurable progress
- DevOps and platform engineers managing autonomous agent workflows
- ML/AI teams running continuous generation or refinement loops
- Development teams using agents for code review, testing, or refactoring at scale
- Anyone deploying agents in production who needs reliability and cost controls
continuous-agent-loop FAQ
Use the Loop Selection Flow: continuous-pr for strict CI/PR control, rfc-dag for RFC decomposition, infinite for exploratory parallel generation, or sequential as the default.
Combine RFC decomposition (ralphinho-rfc-pipeline), quality gates (plankton-code-quality + /quality-gate), eval loop (eval-harness), and session persistence (nanoclaw-repl).
Set up quality gates with measurable acceptance criteria, use eval harnesses to validate progress, and monitor for repeated retries with the same root cause.
Freeze the loop, run /harness-audit to diagnose, reduce scope to the failing unit, and replay with explicit acceptance criteria.
Yes, v1.8+ maintains compatibility with the older autonomous-loops skill while providing the canonical continuous-agent-loop pattern.
Full instructions (SKILL.md)
Source of truth, from affaan-m/ecc.
name: continuous-agent-loop description: Patterns for continuous autonomous agent loops with quality gates, evals, and recovery controls. metadata: origin: ECC
Continuous Agent Loop
This is the v1.8+ canonical loop skill name. It supersedes autonomous-loops while keeping compatibility for one release.
Loop Selection Flow
Start
|
+-- Need strict CI/PR control? -- yes --> continuous-pr
|
+-- Need RFC decomposition? -- yes --> rfc-dag
|
+-- Need exploratory parallel generation? -- yes --> infinite
|
+-- default --> sequential
Combined Pattern
Recommended production stack:
- RFC decomposition (
ralphinho-rfc-pipeline) - quality gates (
plankton-code-quality+/quality-gate) - eval loop (
eval-harness) - session persistence (
nanoclaw-repl)
Failure Modes
- loop churn without measurable progress
- repeated retries with same root cause
- merge queue stalls
- cost drift from unbounded escalation
Recovery
- freeze loop
- run
/harness-audit - reduce scope to failing unit
- replay with explicit acceptance criteria
Related skills
More from affaan-m/ecc and the wider catalog.
continuous-learning
[DEPRECATED] Legacy v1 stop-hook skill extractor—use continuous-learning-v2 instead.
contract-first
Agent skill from affaan-m/ecc.
cost-aware-llm-pipeline
Route models by task complexity, track costs immutably, retry safely, and cache prompts to optimize LLM API spend.
cost-tracking
Track Claude Code token usage, spending, and budgets from local metrics logs.
council
Convene four independent voices—Architect, Skeptic, Pragmatist, Critic—to surface structured disagreement on ambiguous decisions.
council-multi-model
Agent skill from affaan-m/ecc.