agentic-engineers

Experiment orchestration framework with traffic allocation, statistical analysis, and early stopping detection. Use to test routing changes, model upgrades, and role assignments with Welch's t-test significance testing.

DEPRECATED — This skill is no longer maintained. Prometheus and Grafana infrastructure was never implemented. Use local JSON metrics analysis instead.

Scaffolds new SPEC-compliant agentic-engineers agents with a single call. Generates SKILL.md frontmatter, test scaffolds (TDD RED-phase), __init__.py, scripts/ layout, and DELEGATE/HANDBACK protocol templates. Use when creating any new automation agent, task handler, or operational tool. Validates role, model, effort, naming, and dependency graphs (circular dep detection) before writing files. Supports dry-run mode for planning without side effects.

Routine maintenance for Codex sessions: monitor queue state, close completed sub-agents, resume or escalate active work, and keep agent capacity available.

Automated cross-validation of protocol queue integrity. Scans all DELEGATEs/HANDBACKs, validates schema compliance, detects cycles, checks rate limits, generates compliance report. Enables self-referential protocol improvements.

Consolidates provider-specific AI costs into unified metrics across Anthropic, OpenAI, Google Gemini, GitHub Copilot, and Ollama. Enables apples-to-apples cost comparison and savings analysis.

Cost Budget Enforcement (COST-001) — enforces per-session, per-hour, and per-day AI spend limits across all agents and models. Blocks task execution when budgets are exceeded and emits structured alerts for operator action. Integrates with model-selection (COST-003) to pre-screen model choices against budget headroom.

Automated documentation-quality monitoring (MONITORING-001). Scans a markdown documentation tree for broken internal links, missing required sections, stale docs, placeholder/TODO leakage, and structural/readability issues. Produces a structured JSON + human-readable report with a configurable health-score gate.

Discovers and analyzes scripts in the repository, scoring them for utility and integration. Produces SYNC_REPORT.md with recommendations for integration, archival, or deletion of each script.

Continuously discover and document agent/sub-agent integration code/docs/info across all harnesses (OpenCode, Copilot, Claude, PI) to prevent drift and keep integrations fresh. Scans integration code, parses KNOWN_KEYS/KNOWN_MODELS, tracks versioning, identifies drift between docs and code, and generates integration-summary.yaml per harness.

Local Model Runtime support (COST-004) — detects a running local Ollama instance, lists locally-available models, and routes tasks to a zero-cost local model when a suitable one exists, falling back to a cloud provider otherwise. Strategic enabler of up to ~95% cost reduction.

Cost-quality optimization agent that analyzes tradeoffs, scores routing candidates, and proposes A/B tests. Use to generate task routing recommendations and experiment designs based on TokenAdvisor findings.

Model Selection Optimization (COST-003) — recommends optimal AI models for tasks given budget constraints, quality targets, and latency requirements. Computes the Pareto cost-quality frontier and supports mixed-model routing simulation.

Harness-native queue polling scheduler. Invokes Orchestrator skill on a recurring schedule to process queued DELEGATEs. Uses environment variables (read live) to detect session ID and harness. Implements queue polling as a SKILL—no external daemons, cron jobs, or background processes needed. Can be re-awakened via skill invocation.

In-harness queue orchestration system that implements the DELEGATE/HANDBACK protocol lifecycle. Manages queue state machine (7 states: incoming, claimed, processing, done, failed, crashed, retry-pending), polls queue for new tasks, spawns sub-agents via Agent tool, correlates HANDBACK results, recovers from crashes, invokes quality gates, and detects idle conditions. Core system that makes DELEGATE/HANDBACK actually work.

Runtime protocol validation for DELEGATEs/HANDBACKs against protocol-core-v1.0.yaml. Validates core fields, extensions, and unknown fields with forward-compatibility support. <5ms validation time.

Atomic queue operations for DELEGATE/HANDBACK lifecycle with cycle detection, rate limiting, and validation. Enables decentralized sub-task creation and reduces orchestrator bottleneck.

Live queue monitoring dashboard (curses TUI) for real-time visibility into DELEGATE/HANDBACK protocol queue status, metrics, and task progress across incoming, processing, and done states.

Local-queue visibility skill — query and inspect the per-session, per-harness filesystem queue by state (incoming backlog, processing orphans to resume, done results/next-steps). Format-agnostic (json + yaml). The local stepping stone toward the external memory-API queue interface.

Auto-sync queue DELEGATEs ↔ TODO.md on task lifecycle events

Meta-skill for automated session transcript analysis. Reads session artifacts (DELEGATEs, HANDBACKs, metrics, conversation history) to detect repetitive patterns, quality anomalies, drift detection, and effort mismatch. Outputs actionable recommendations for skill enhancement and process improvements.

Canonical pattern definition for the skill self-improvement feedback loop. Defines how agents emit structured skill_feedback in HANDBACKs, what fields are expected, and how feedback accumulates into improvement tasks. Every other skill references this as the single source of truth for its ## Self-Improvement section.

Exclusive SPEC.md change protection with structured proposal interface, impact analysis, multi-level authorization, immutable audit trail, and rollback capability. Only Principal/Security/Lead Engineers can invoke. Enforces all SPEC.md modifications through proposal→analysis→approval→changelog workflow with tamper-evident audit trail.

Validates implementation compliance with SPEC.md. Parses SPEC.md requirements and constraints, analyzes git diffs, detects violations and rollbacks, and generates compliance reports. Use as a pre-merge gate or post-merge audit to prevent spec drift.

Real-time and historical token usage capture, analysis, and forecasting skill for agents

Validates end-to-end delegation workflows for correctness. Generates ASCII workflow diagrams showing agent data flow. Checks for missing quality gates, incomplete feedback loops, blocked exit criteria, and unused components.

DEPRECATED — merged into harness-integration-tracker as scripts/opencode_sync.py (Wave 3, m3-skills-deprecation, 2026-06-14). Use harness-integration-tracker instead.

[DISABLED] Initializes new repositories with the agentic-engineers framework. Currently disabled pending further discussion about repo modification policies.

DEPRECATED — merged into agent-creator (Wave 3, m3-skills-deprecation, 2026-06-14). Use agent-creator --type skill instead. Create new agentic-engineers skills following the agentskills.io specification.

DEPRECATED — merged into usage-tracking as role-analysis sub-command (Wave 3, m3-skills-deprecation, 2026-06-14). Use usage-tracking/scripts/role_analysis.py instead.

Validates test fixture synchronization with code changes. Detects orphaned test expectations, stale fixtures, and missing test coverage for code updates. Use as pre-merge gate to catch test-code drift before regressions reach CI.

Session-scoped, harness-scoped queue path isolation for multi-harness agentic-engineers workflows. Ensures Claude, Copilot, GPT, and local agents never collide on the same queue directories.

Runtime enforcement of canonical queue paths for DELEGATE/HANDBACK files. Enforces canonical path: ~/.agentic-engineers/{harness}/{session-id}/queue/ Rejects legacy paths and injection attacks. Integrates with git hooks to prevent non-canonical paths in commits.

Disciplined queue-to-branch-to-PR-to-merge workflow for code quality and CI stability

Comprehensive checklist ensuring all new features are integrated into the framework correctly

Comprehensive harness integration test suite for catching compatibility flaps and feature regressions early

<Brief description of what this skill does and when to invoke it. Should be ≤200 characters for clean harness rendering.>

Tri-level verification of AGENTS.md against SPEC.md. Blocks model downgrade attacks and unauthorized agent definitions.

Validate security-critical DELEGATE fields and enforce routing rules. Prevents security work from being mis-routed to general engineers.

Efficiently monitor GitHub Actions runs and extract review feedback. Uses adaptive polling intervals (30s for initial/active, 60s for stabilized) to minimize API calls while maintaining responsive feedback. Extracts logs, identifies failures, and documents Copilot review comments.