Skip to content
/
Tier A A4 Markdown scaffold

Cursor Agentic Development Pipeline

cursor-agentic-pipeline · ChrisThompsonTLDR/agentic-programming · ★ 12 · last commit 2026-05-21

Automate building an Obsidian knowledge base of Laravel packages, skills, and MCP servers using GitHub Copilot cloud agents.

Best whenEach agent invocation writes exactly one file — no side effects, no batch operations, surgical output only.
Skip ifRelying on cached knowledge instead of fetching fresh data, Guessing field values instead of leaving them blank
vs seeds
taskmaster-ai's slash-command/file-write paradigm, but taskmaster targets code task management while this framework targets research …
Primitive shape 12 total
Commands 3 Skills 3 Subagents 3 MCP tools 3
00

Summary

Cursor Agentic Development Pipeline — Summary

agentic-programming by ChrisThompsonTLDR is a collection of GitHub Copilot Cloud Agents — custom .github/agents/*.agent.md definitions that each research a specific domain and write Obsidian-style structured notes to the repository. Three agents ship: laravel-package (researches Laravel packages and writes .steering/laravel-packages/<vendor>__<package>.md), skill-research (researches GitHub-hosted skills and writes .steering/skills/ notes), and mcp-research (researches MCP servers and writes mcps/ notes). Each agent uses external MCP servers (DeepWiki, Context7, GitHub) for research, and the repository memory feature lets agents learn from past runs. A GitHub Actions workflow auto-syncs template files. The framework is primarily a documentation-generation pipeline for a Laravel PHP developer's knowledge base, not a general-purpose coding agent system. It differs from all 11 seeds by targeting GitHub Copilot's cloud agent runtime (.github/agents/) rather than local IDE tools, and by focusing on knowledge acquisition/research rather than code generation.

01

Overview

Cursor Agentic Development Pipeline — Overview

Origin

Created by ChrisThompsonTLDR (GitHub user), actively maintained as of May 2026. The repository is titled "Agentic Programming" and focuses on GitHub Copilot Cloud Agents for research and documentation. Despite the batch assignment slug cursor-agentic-pipeline, the repository uses GitHub Copilot's agent system (.github/agents/), not Cursor's agent system.

Philosophy

The README states the framework is "a collection of custom GitHub Copilot Cloud Agents for research and documentation." The design philosophy centers on:

  1. Research automation — agents autonomously gather information from authoritative sources (GitHub, Packagist, Laravel News, DeepWiki, Context7) and write structured notes
  2. Obsidian-style knowledge base — output files use Obsidian-compatible YAML frontmatter and wiki links ([[Laravel Packages]])
  3. Incremental memory — GitHub Copilot's memory feature stores learnings from past agent runs for formatting consistency
  4. Minimal user effort — user types one command like /laravel-package spatie/laravel-markdown-response; the agent handles all research and file writing

Target Use Case

A Laravel PHP developer who wants to maintain a structured, AI-queryable knowledge base of packages, skills, and MCP servers they use. The .steering/ directory convention is specific to GitHub Copilot's Memory and "steering documents" feature.

Key Design Opinions

  1. One command, one output file — each agent writes exactly one file per invocation ("Surgical Output" principle quoted in the agent prompt)
  2. Field verification over guessing — every frontmatter field must be verified from a specific data source before writing
  3. Research breadth — agents use multiple MCP servers in parallel (DeepWiki for semantic docs, Context7 for live docs, GitHub MCP for repo stats)
  4. Conventional naming — output files use <vendor>__<package>.md double-underscore convention for filesystem-safe names

Manifesto-Style Quote (from laravel-package.agent.md)

"Think Before Acting — Parse the vendor/package slug first. If it is ambiguous or malformed, stop and ask."

"Simplicity First — Populate only fields you can verify from sources. Leave fields blank rather than guessing."

"Surgical Output — Write exactly one file. Do not create, modify, or delete anything else."

"Goal-Driven — Success = a valid .md file at the correct path with all resolvable fields filled in."

02

Architecture

Cursor Agentic Development Pipeline — Architecture

Distribution Type

template-bundle — repository is cloned or forked; agents live in .github/agents/ and are invoked via GitHub Copilot Chat.

Install

1. Fork or clone the repository
2. Go to Repo Settings > Secrets and variables > Copilot env and add:
   - DEEPWIKI_KEY (from api.deepwiki.com)
   - CONTEXT7_KEY (from context7.com)
3. Go to Repo Settings > Copilot > Memory and enable it
4. Go to Repo Settings > Copilot > MCP and add the MCP config from .github/mcp.json
5. Use in GitHub Copilot Chat: /laravel-package spatie/laravel-markdown-response

Directory Tree

agentic-programming/
├── .github/
│   ├── agents/
│   │   ├── laravel-package.agent.md       # Laravel package research agent
│   │   ├── mcp-research.agent.md          # MCP server research agent
│   │   └── skill-research.agent.md        # Skill/framework research agent
│   ├── skills/
│   │   ├── laravel-research/
│   │   │   ├── SKILL.md                   # Note template generator skill
│   │   │   └── laravel-package-template.md # Blank template reference
│   │   ├── skill-research/
│   │   │   ├── SKILL.md                   # Skill note generator
│   │   │   └── skill-template.md          # Blank skill template
│   │   └── agentic-programming/
│   │       └── README.md                  # Skill docs
│   ├── mcp.json                           # MCP server config (Context7 + DeepWiki)
│   └── workflows/
│       └── sync-skill-template.yml        # Auto-syncs templates/skill.md
├── .openclaw/
│   └── github-review-digest-state.json   # State file for review digest
├── .steering/
│   ├── laravel-packages/                  # Output: Laravel package notes
│   └── skills/                           # Output: skill/framework notes
├── mcps/                                  # Output: MCP server notes
└── README.md

Required Runtime

  • GitHub Copilot subscription (for GitHub Copilot Cloud Agents)
  • GitHub repository with Copilot enabled
  • DeepWiki API key
  • Context7 API key

Target AI Tools

  • GitHub Copilot (primary — cloud agent runtime)
  • Not compatible with local IDE agents (Cursor, Roo Code, Claude Code)

Config Files

  • .github/mcp.json — MCP server config (Context7 HTTP, DeepWiki HTTP)
  • GitHub repo secrets: DEEPWIKI_KEY, CONTEXT7_KEY
  • GitHub Copilot Memory: enabled via repo settings

MCP Config (.github/mcp.json)

{
  "mcpServers": {
    "Context7": {
      "type": "http",
      "tools": ["*"],
      "url": "https://mcp.context7.com/mcp"
    },
    "deepwiki": {
      "type": "http",
      "tools": ["*"],
      "url": "https://mcp.deepwiki.com/mcp"
    }
  }
}
03

Components

Cursor Agentic Development Pipeline — Components

Agents (.github/agents/)

File Trigger Output Path
laravel-package.agent.md /laravel-package <vendor/package> .steering/laravel-packages/<vendor>__<package>.md
skill-research.agent.md /skill-research <github-url-to-skill> .steering/skills/<owner>__<skill-name>.md
mcp-research.agent.md /mcp-research <url> mcps/<owner>__<name>.md

Skills (.github/skills/)

Skill Purpose
laravel-research/SKILL.md Template for generating Laravel package notes; defines YAML frontmatter schema and field-to-source mapping
skill-research/SKILL.md Template for skill/framework research notes
agentic-programming/README.md Documentation skill

MCP Servers (via .github/mcp.json)

Server URL Purpose
Context7 https://mcp.context7.com/mcp Live library documentation lookup
DeepWiki https://mcp.deepwiki.com/mcp Semantic repository documentation

Each agent also uses the GitHub MCP server (defined inline in the agent's mcp-servers: frontmatter, using repo-level GitHub MCP SSE).

GitHub Actions Workflows

Workflow Trigger Purpose
sync-skill-template.yml Push to templates/ Copies templates/skill.md → skill-research/skill-template.md

State Files

File Purpose
.openclaw/github-review-digest-state.json State for a review digest feature (not documented in README)

Templates

File Purpose
.github/skills/laravel-research/laravel-package-template.md Blank template reference for Laravel package notes
.github/skills/skill-research/skill-template.md Blank template for skill research notes

Total Primitive Count

  • Agents: 3
  • Skills: 3
  • MCP servers: 2 (global) + 1 GitHub MCP (per-agent)
  • Workflows: 1
  • Templates: 2
05

Prompts

Cursor Agentic Development Pipeline — Prompts

Excerpt 1: laravel-package Agent System Prompt (laravel-package.agent.md)

---
name: laravel-package
description: "Researches Laravel packages and generates Obsidian notes. Usage: /laravel-package <vendor/package>"
tools: ["read", "search", "edit", "browser"]
mcp-servers:
  deepwiki:
    type: sse
    url: https://api.deepwiki.com/sse
    tools: ["*"]
    headers:
      Authorization: "Bearer $DEEPWIKI_KEY"
  context7:
    command: npx
    args: ["@context7/mcp-server"]
    tools: ["*"]
    env:
      CONTEXT7_API_KEY: $CONTEXT7_KEY
  github:
    type: sse
    url: https://mcp.github.com/sse
    tools: ["*"]
---

You are a Laravel package researcher. For `/laravel-package <vendor/package>`:

## Core Principles

**Think Before Acting** — Parse the vendor/package slug first. If it is ambiguous or malformed, stop and ask.

**Simplicity First** — Populate only fields you can verify from sources. Leave fields blank rather than guessing.

**Surgical Output** — Write exactly one file. Do not create, modify, or delete anything else.

**Goal-Driven** — Success = a valid `.md` file at the correct path with all resolvable fields filled in.

## Steps

1. Parse `<vendor/package>` (e.g., `spatie/laravel-markdown-response`).
2. Research — fetch fresh data from each source below. Do not rely on cached knowledge.
3. Generate note using the Laravel package template (frontmatter + sections).
4. Write to `.steering/laravel-packages/<vendor>__<package>.md`.
5. Leverage repo Memories for consistent formatting patterns.

Prompting technique: YAML frontmatter in .agent.md declares tools and MCP servers (Copilot cloud agent format). System prompt uses bold-header principles ("Think Before Acting", "Simplicity First") followed by a numbered execution plan. The four principles are explicitly labeled with a name — unusual compared to most frameworks which enumerate rules without names. This mirrors the Agent Operating Standard pattern.

Excerpt 2: laravel-research Skill (SKILL.md) — Template Schema

---
name: laravel-research
description: Generates Laravel package notes from the Laravel package template. Use for structured output.
---
You are the Laravel Package Note Generator.

Use this EXACT template for output:

```markdown
org: VENDOR
package: package-name
github_url: https://github.com/VENDOR/package-name
docs_url: 
composer_require: composer require VENDOR/package-name
author: 
announce_date: 
latest_release: vX.Y.Z (DATE)
...
tags: [laravel, packages, FEATURE1, FEATURE2]
---

# PACKAGE_NAME

**DESCRIPTION**

## Key Features
- FEATURE1
- FEATURE2
...


**Prompting technique**: Template injection — the skill is a system prompt that specifies the exact output format. The `Use this EXACT template` instruction forces structural fidelity. The skill exists separately from the agent to allow reuse across multiple agents.

---

## Excerpt 3: Field Resolution Rules Table

```markdown
## Field Resolution Rules

| Field | Source |
|-------|--------|
| `author` | Packagist JSON (`packagist.org/packages/<v>/<p>.json`) → `versions.dev-main.authors[0].name` |
| `stars` | GitHub MCP `search_repositories` or `get_repository` → `stargazers_count` |
| `latest_release` | GitHub MCP `get_latest_release` → format `vX.Y.Z (YYYY-MM-DD)` |
| `release_date` | GitHub MCP `get_latest_release` → `published_at` (YYYY-MM-DD) |
| `downloads_30d` | Packagist stats page → "Last 30 days" installs |
| `announce_date` | Laravel News article publication date |
| `laravel_news_url` | Search `laravel-news.com <package>` and verify URL |
| `docs_url` | GitHub repo `homepage` field or README docs link |

Prompting technique: Explicit source-to-field mapping table. Prevents hallucination by requiring each field to be resolved from a named, verifiable source rather than inferred from training data.

09

Uniqueness

Cursor Agentic Development Pipeline — Uniqueness

Differs From Seeds

This framework differs from all 11 seeds in targeting GitHub Copilot's cloud agent runtime (.github/agents/*.agent.md) rather than local IDE tools. No seed uses this runtime. The closest structural analogs are:

  • taskmaster-ai (seed): also uses a slash-command invocation model and writes to structured output files; but taskmaster runs as an MCP server in Cursor/Claude Code, while this framework runs as a GitHub Copilot cloud agent
  • openspec (seed): also produces structured markdown artifacts from agent invocations; but openspec targets code planning workflows while this framework targets research documentation

The framework's unique contribution is the Obsidian knowledge base as the output layer: every agent writes to a versioned .steering/ or mcps/ directory in Obsidian-compatible format, creating a queryable, cross-linkable knowledge graph of researched dependencies and tools.

The laravel-research/SKILL.md pattern — a skill that is a template schema enforcer rather than a behavioral instruction — is distinct from all seeds. The skill exists to prevent format drift across agent invocations.

Positioning

  • Niche: Laravel PHP developers wanting automated package/skill/MCP research notes in Obsidian format
  • Cross-tool portability: extremely low — only works with GitHub Copilot cloud agents; the .github/agents/ format is not used by any other tool
  • Scalability pattern: scales horizontally (run agent 100 times to build a 100-package knowledge base) rather than vertically (one complex agent)

Naming Alert

The batch-assigned slug cursor-agentic-pipeline is misleading: this framework does not use Cursor at all. It uses GitHub Copilot. The repository name is agentic-programming.

Observable Failure Modes

  1. MCP authentication failures — two external API keys required; any key expiry breaks research quality
  2. Packagist scraping fragility — the downloads_30d field requires scraping a stats page; layout changes break it
  3. Laravel-specific utility — three of the three agents are PHP/Laravel-specific; minimal reuse value for non-Laravel developers
  4. Copilot Memory opacity — GitHub Copilot Memory is a black box; exact formatting preferences stored are not inspectable
  5. GitHub Copilot dependency — requires active Copilot subscription; not usable in offline or open-source CI environments

Cross-References

  • No relation to Cursor despite the batch slug name
  • The "Surgical Output" and "Think Before Acting" principles in the agent prompt appear to be influenced by Anthropic's Claude Operating Standard (or a similar agent design philosophy)
04

Workflow

Cursor Agentic Development Pipeline — Workflow

Laravel Package Research Workflow

Step Action Artifact
1 User types /laravel-package spatie/laravel-markdown-response Command trigger
2 Agent parses <vendor/package> slug (validation)
3 Agent fetches: GitHub MCP stats, Packagist JSON, Laravel News search, Context7/DeepWiki docs Research data
4 Agent generates note using laravel-research skill template YAML frontmatter + body
5 Agent writes .steering/laravel-packages/spatie__laravel-markdown-response.md Research note
6 GitHub Copilot Memory stores formatting patterns for future runs Memory update

Field Resolution Protocol

The agent must verify each frontmatter field from a specific authoritative source before writing:

Field Required Source
author Packagist JSON versions.dev-main.authors[0].name
stars GitHub MCP stargazers_count
latest_release GitHub MCP get_latest_release → vX.Y.Z (YYYY-MM-DD)
downloads_30d Packagist stats page "Last 30 days"
announce_date Laravel News article publication date
laravel_news_url Searched and verified URL

Approval Gates

None. Agents are fully autonomous once triggered. The --completion-promise pattern from other frameworks does not apply here. If a slug is ambiguous or malformed, the agent stops and asks.

Output Naming Convention

<vendor>__<package>.md — double underscore separator ensures filesystem compatibility for vendors with hyphens.

Skill Research Workflow

Analogous to Laravel package workflow but targets GitHub-hosted skill repos:

  • Trigger: /skill-research <github-url-to-skill>
  • Output: .steering/skills/<owner>__<skill-name>.md

MCP Research Workflow

  • Trigger: /mcp-research <url>
  • Output: mcps/<owner>__<name>.md

Phase-to-Artifact Map

Phase Artifact
Invocation Trigger command
Research (parallel MCP calls) Raw data from GitHub/Packagist/DeepWiki/Context7
Note generation YAML frontmatter + structured Markdown body
File write .steering/ or mcps/ .md file
Memory update GitHub Copilot Memory (formatting patterns)
06

Memory Context

Cursor Agentic Development Pipeline — Memory & Context

Primary Memory Mechanism

GitHub Copilot Memory — a GitHub-native feature where the Copilot service stores learnings from agent interactions. The framework explicitly requires enabling this:

"Go to Repo Settings > Copilot > Memory and enable it so the agent learns from past runs."

Memory is used to maintain consistent note formatting across runs.

State Files

File Purpose
.openclaw/github-review-digest-state.json State for a review digest workflow (contents not fully documented in README)

Output as Persistent Knowledge Base

The .steering/ and mcps/ directories are the persistent knowledge base:

.steering/
├── laravel-packages/     # One .md file per researched package
└── skills/               # One .md file per researched skill
mcps/                     # One .md file per researched MCP server

Each note includes YAML frontmatter with metadata and Markdown body with structured content. The Obsidian-style wiki links ([[Laravel Packages]]) create a navigable knowledge graph if used with Obsidian.

Cross-Session Continuity

GitHub Copilot Memory provides cross-session continuity for agent formatting preferences. The generated .steering/ files provide cross-session continuity for the knowledge base content.

Compaction

Not applicable. Knowledge base grows incrementally.

Context Handoff

No formal handoff mechanism. Each agent invocation is independent. Memory provides implicit cross-run context.

Memory Persistence

Hybrid:

  • GitHub Copilot Memory (cloud-managed, repo-scoped)
  • File-based .steering/ and mcps/ directories
07

Orchestration

Cursor Agentic Development Pipeline — Orchestration

Multi-Agent Pattern

Each agent invocation is single-agent. However, within an agent run, parallel MCP calls are made to multiple data sources simultaneously (GitHub, DeepWiki, Context7, Packagist). This is intra-agent parallelism, not multi-agent orchestration.

Pattern: none at the agent level; parallel-fan-out within each agent for MCP tool calls.

Execution Mode

one-shot — each /command invocation runs one agent to completion, writes one file, and terminates.

Isolation

None. Agents write directly to the working tree.

Multi-Model Routing

None defined. GitHub Copilot's model selection is used.

Prompt Chaining

Weak: the laravel-research and skill-research SKILL.md files are loaded by their respective agents as context templates. The agent system prompt + the skill template form a two-stage prompt chain.

Agent Coordination

None. Agents are independent. If the user researches 10 packages, they invoke the agent 10 times sequentially.

MCP Tool Parallelism

Each agent's research phase fetches from multiple sources. The agent prompt says "Do not rely on cached knowledge" — requiring fresh fetches every run.

MCP servers per agent:

  • laravel-package: DeepWiki (SSE) + Context7 (stdio) + GitHub (SSE)
  • skill-research: DeepWiki + Context7 + GitHub
  • mcp-research: DeepWiki + Context7 + GitHub

Consensus Mechanism

None.

Subagent Spawn

None. GitHub Copilot cloud agent system does not support spawning sub-agents in the same way as Cursor's SDK.

08

Ui Cli Surface

Cursor Agentic Development Pipeline — UI / CLI Surface

Dedicated CLI Binary

None.

Local Web Dashboard

None.

Primary Interface

GitHub Copilot Chat (in GitHub.com, VS Code, or JetBrains):

  • User selects the named agent from the agent picker
  • Types a slash command with arguments: /laravel-package spatie/laravel-markdown-response
  • Agent runs autonomously in GitHub Copilot's cloud infrastructure

GitHub Integration Points

Integration Purpose
GitHub Actions (sync-skill-template.yml) Auto-syncs template files on push
GitHub Copilot Memory Stores formatting preferences
GitHub Copilot MCP Config Loads Context7 + DeepWiki MCP servers
GitHub Copilot Secrets Stores DEEPWIKI_KEY + CONTEXT7_KEY

IDE Support

The agents run in:

  • GitHub Copilot on GitHub.com
  • GitHub Copilot in VS Code
  • GitHub Copilot in JetBrains IDEs

They do not run in Cursor, Roo Code, or Claude Code.

Observability

  • Each agent writes one .md file — the file itself is the output artifact
  • GitHub Copilot chat history shows the agent's research steps
  • No dedicated log files or audit trail beyond git history of written files

Naming Convention

Output files use <owner>__<name>.md (double underscore) — intended for Obsidian compatibility and filesystem safety.

Related frameworks

same archetype · same primary tool · same memory type

BMAD-METHOD ★ 48k
A4 Markdown scaffold

Provides a full agile delivery lifecycle with named expert-persona AI collaborators that elicit the human's best thinking rather…

Agent OS ★ 4.6k
A4 Markdown scaffold

Extracts implicit codebase conventions into token-efficient markdown standards files and injects them selectively into AI agent…

Claude Conductor ★ 367
A4 Markdown scaffold

Gives Claude Code a persistent, cross-linked, auto-analyzed documentation system so it retains codebase context across sessions.

Spec-Driver (Greenfield Spec-Driven Development) ★ 25
A4 Markdown scaffold

Prevents spec rot in AI-assisted development by making implementation changes flow back into evergreen, authoritative specs via…

Anthropic Knowledge Work Plugins ★ 16k
A4 Markdown scaffold

Role-specialized plugin bundles with live MCP connectors that turn Claude into a domain expert for enterprise knowledge workers.

Codex Integration for Claude Code (skill-codex) ★ 1.3k
A4 Markdown scaffold

Single Claude Code skill that handles Codex CLI invocation correctly (stdin blocking, thinking token suppression, session resume)…