Agents

Jared's AI Team

[](https://github.com/JaredLin1217/Agents/actions/workflows/checkpoint.yml) [](https://github.com/JaredLin1217/Agents/actions/workflows/public-updates.yml)

Jared's AI Team is a repo-local AI Agents workflow system for Codex-style engineering sessions. It provides compact routing rules, project-local operating policy, deployable governance files, runtime evidence boundaries, multi-agent coordination contracts, official-docs-first foundations, freshness-bound durable memory, change-aware verification, and release packaging checks.

The repository is intentionally small and source-neutral. It is designed to be deployed into target projects without copying local runtime state, live Codex configuration, private thread identifiers, approval scratch files, or provider repo history.

Current Version

Current Agents workflow version: 2.8.0 (precision-efficiency).

Canonical version source: docs/agents/version.yaml

Version positioning: 2.8.0 is the latest-only precision release. It uses progressive disclosure, one state owner, freshness-bound durable memory, change-aware verification, and current official foundations to reduce routine context and verification cost without weakening release evidence.

What This Project Provides

Compact AI Runtime Routing

docs/agents/ai-runtime.yaml is the first routing layer after AGENTS.md. It keeps default context small and expands only the named canonical YAML files needed for the current task. This keeps ordinary turns fast while still allowing deeper routes for deployment, multi-agent work, runtime evidence, provider adapters, route packs, knowledge footprints, and foundation creation.

Foundation Creation Governance

The docs/agents/*.yaml files define the project-owned operating standard: policy, verification, deployment, organization, model tiering, dispatch, schemas, runtime execution evidence, collaborator windows, context compaction, foundation creation, and official OpenAI capability boundaries.

docs/agents/openai-foundations.yaml converts official OpenAI documentation into project-native primitives for Structured Outputs, conversation state, agent handoffs, Codex instructions, skills, memories, subagents, prompt caching, compaction, token counting, predicted outputs, and evaluation-backed release decisions.

Durable Memory With Freshness

docs/memory/index.md is the compact recall index. Detailed entries are loaded only when the current route needs them, are capped by the context budget, and must carry source references, confidence, status, and review dates. Current repository evidence always overrides remembered state; checked-in memory never overrides canonical policy.

Runtime Evidence Without Repo Pollution

Runtime evidence is local by design. Files such as .agents/runtime/**, .workflow/**, local Codex configuration, status files, temporary approval records, and live collaborator state are ignored or blocklisted from release packages and deployments unless a task explicitly authorizes a targeted runtime operation.

Deployable Workflow Package

The deployment system can install the AI Agents workflow into another authorized repository using an allowlisted file set. It supports dry-run first operation, target layout detection, target-owned state preservation, source neutral template checks, and release package export.

Multi-Agent and Enterprise Dispatch Contracts

The workflow includes optional two-layer dispatch for controller, department leaders, and workers. It defines assignment fields, reports, scoring, reconciliation, cleanup evidence, collaborator windows, and escalation boundaries. Runtime multi-agent claims require live employee evidence.

Repository Map

| Path | Purpose | |---|---| | AGENTS.md | Repo-local session rules and closeout contract. | | .agents/skills/project-isolation-workflow/ | Project-local skill for isolation, memory, deploy, agents, handoff, skills, and maintenance. | | docs/agents/ | Canonical AI Agents workflow YAML files. | | docs/runbooks/ | Human-readable operating procedures. | | docs/templates/agents/ | Source-neutral deployable template bundle. | | schemas/ | Lightweight schema contracts used by validation. | | scripts/ | Validation, deployment, release, route-pack, runtime, and cleanup helpers. | | tests/agents-governance-fixtures/ | Validation fixtures for schema and workflow contracts. | | artifacts/ | Provider-side audit and evaluation artifacts. |

Read Order

1. AGENTS.md
2. docs/agents/ai-runtime.yaml
3. Route-specific canonical YAML named by ai-runtime.yaml
4. docs/agents/verify.yaml before state claims, edits, commits, deployment, or release
5. docs/project-structure.md for ownership and deployability boundaries

Do not use docs/templates/agents/** as the active rule source. Templates are deployment material, not the current repo's live operating authority.

Core Commands

Change-aware validation for routine work:

.\scripts\validate-changes.ps1 -Profile Auto -Explain

The auto profile selects Fast, Policy, or Full from changed-path risk. Use an explicit profile when the claim scope requires stronger evidence.

Full release readiness and quality score:

.\scripts\validate.ps1 -Full -Score

Capture sanitized v2.8.0 practice runtime release evidence:

.\scripts\capture-runtime-evidence.ps1 -OutputPath .\docs\evidence\releases\v2.8.0-runtime-evidence.json -Full -Practice -Quiet

Deployment dry-run into an authorized target:

.\scripts\deploy-agents-workflow.ps1 -TargetPath "D:\target\repo" -Mode core_bootstrap -LayoutProfile auto -DryRun

Upgrade an authorized target after dry-run review:

.\scripts\deploy-agents-workflow.ps1 -TargetPath "D:\target\repo" -Mode core_bootstrap -LayoutProfile auto -Upgrade

Runtime execution evidence example:

.\scripts\agents-runtime.ps1 -Action NewRun -RunId "example"
.\scripts\agents-runtime.ps1 -Action AddStep -RunId "example" -Step "read_only"
.\scripts\agents-runtime.ps1 -Action AddResult -RunId "example" -Result "completed" -Summary "read-only example completed"
.\scripts\agents-runtime.ps1 -Action Verify -RunId "example"
.\scripts\agents-runtime.ps1 -Action Cleanup -RunId "example"

Closed subagent residue verification:

.\scripts\agents-cleanup.ps1 -Action Verify -RuntimeIds "" -ParentThreadId ""
.\scripts\agents-cleanup.ps1 -Action Cleanup -RuntimeIds "" -ParentThreadId "" -Force

Deterministic route-pack export:

.\scripts\export-route-pack.ps1 -RouteId core_system

Release package export:

.\scripts\export-release-package.ps1

Validation Coverage

The main validator covers:

• Lightweight YAML syntax and required canonical files
• AI runtime compact routing and expand-only route behavior
• Canonical schema contract checks
• Change-aware validation profile selection and changed-file syntax checks
• Memory index, metadata, freshness, source-reference, and template integrity
• Enterprise dispatch, workflow artifacts, context compaction, collaborator windows, and core runtime integrity
• Runtime/local boundary checks
• Durable English-only and placeholder gates
• Exact-pair drift checks for deployable mirrors
• Deployment manifest, blocklist, source-neutral template, and self-test checks
• Multi-agent workflow integrity and cleanup evidence contracts
• CI workflow stability
• P0-P5 readiness evidence
• Size gates for routing files, scripts, project skill, and tracked repository footprint
• Release package export checks

Operating Boundaries

• Runtime state is local evidence, not deployable project source.
• Target deployments preserve target-owned state by default.
• Hard-isolation claims require current runtime, tool, OS, account, or cloud

enforcement evidence.

• Runtime multi-agent claims require live employee evidence and cleanup

reconciliation.

• OpenAI API, Apps SDK, Codex, Agents SDK, model, and tool guidance should

use official OpenAI developer documentation first.

• Global memory and global/system skills are off by default unless explicitly

requested.

Quality Target

The project aims for a small, portable, evidence-backed AI Agents operating standard:

• Compact default context
• Clear route expansion
• Source-neutral deployment
• Runtime evidence without repository pollution
• Repeatable validation and release checks
• Explicit claim boundaries for multi-agent execution and hard isolation
• Schema-first feature creation from official OpenAI capabilities
• Latency, cost, fallback, and evaluation gates before release

License

Copyright 2026 Yu-Jie, Lin.

Licensed under the Apache License, Version 2.0. See LICENSE for the full license text and NOTICE for the project notice and additional disclaimer.

Unless required by applicable law or agreed to in writing, this repository is provided on an "AS IS" basis, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.

Source & license

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

• Author: JaredLin1217
• Source: JaredLin1217/Agents
• License: Apache-2.0

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