Skip to content

INSTALLATION.md

Latest commit

 

History

History
355 lines (250 loc) · 9.22 KB

INSTALLATION.md

File metadata and controls

355 lines (250 loc) · 9.22 KB

Installation Guide

Progressive Estimation works with any AI coding client that supports custom instructions. This guide covers installation for each major client.

Note

Progressive loading (only loading what's needed per phase) is fully supported in Claude Code, Cursor, and GitHub Copilot. For other clients, you may need to consolidate the skill into a single file or accept that all content loads at once. See the notes per client below.

Claude Code

Full progressive loading support. The skill was designed for Claude Code and works best here — files load on demand per phase.

# Clone into your skills directory
git clone https://github.com/Enreign/agent-skills.git ~/.claude/skills/agent-skills

That's it. The skill auto-triggers when you ask for estimates. You can also invoke it explicitly with /progressive-estimation.

How it works: Claude Code reads SKILL.md as the entry point and loads reference files on demand as the estimation workflow progresses through phases.

Cursor

Supports conditional loading via glob and agent-decided rules.

Option A: Agent-Decided Rule (Recommended)

Create a single rule file that Cursor loads when estimation is relevant:

# Create the rules directory
mkdir -p .cursor/rules

# Copy the skill file
cp path/to/progressive-estimation/SKILL.md .cursor/rules/progressive-estimation.mdc

Add this frontmatter to the top of the .mdc file:

---
description: "Use when the user asks for time estimates, effort sizing, story points, or estimation of development tasks"
alwaysApply: false
---

Option B: Full Skill with References

For the complete skill with all reference files:

# Copy the entire skill into your project
cp -r path/to/progressive-estimation .cursor/rules/progressive-estimation

Create .cursor/rules/progressive-estimation.mdc:

---
description: "Use when the user asks for time estimates, effort sizing, or estimation"
alwaysApply: false
---
For estimation tasks, read and follow the workflow in:
@.cursor/rules/progressive-estimation/SKILL.md

Reference files are in @.cursor/rules/progressive-estimation/references/

Note

Cursor does not have a documented character limit per rule, but keep the main rule file concise and let @file references pull in details.

Windsurf

Partial progressive loading via Model Decision mode.

# Create the rules directory
mkdir -p .windsurf/rules

# Copy the main skill file
cp path/to/progressive-estimation/SKILL.md .windsurf/rules/progressive-estimation.md

Warning

Windsurf limits individual rule files to 6,000 characters and combined rules to 12,000 characters total. The full SKILL.md fits within this limit, but reference files will need to be loaded via chat context rather than as rules.

To use the full skill, keep SKILL.md as the rule and add reference files to your project directory. Ask the AI to read them when needed:

Read references/formulas.md and estimate this task: ...

GitHub Copilot

Supports path-specific progressive loading.

Option A: Repository-Wide Instructions

# Create the instructions directory
mkdir -p .github/instructions

# Copy the main skill file
cp path/to/progressive-estimation/SKILL.md .github/copilot-instructions.md

Or for conditional loading:

cp path/to/progressive-estimation/SKILL.md .github/instructions/estimation.instructions.md

Add frontmatter for conditional activation:

---
applyTo: "**/*"
---

Option B: Full Skill with References

Copy the entire skill into your repo and reference it:

cp -r path/to/progressive-estimation ./progressive-estimation

In .github/copilot-instructions.md, add:

For estimation tasks, follow the workflow in [SKILL.md](progressive-estimation/SKILL.md).
Reference files are in [references/](progressive-estimation/references/).

Note

GitHub Copilot supports #file:path/to/file syntax to reference specific files within instruction files.

Cline (VS Code)

No progressive loading — all rules load together.

Option A: Single File

# Copy to project root
cp path/to/progressive-estimation/SKILL.md .clinerules

Option B: Directory with Multiple Files

# Create rules directory
mkdir -p .clinerules

# Copy skill and key references
cp path/to/progressive-estimation/SKILL.md .clinerules/01-estimation-workflow.md
cp path/to/progressive-estimation/references/formulas.md .clinerules/02-estimation-formulas.md
cp path/to/progressive-estimation/references/questionnaire.md .clinerules/03-estimation-questionnaire.md

Note

All files in .clinerules/ load together. Number-prefix the files to control loading order. Only include the references you use most often to avoid consuming too much context.

Option C: Global Custom Instructions

In VS Code: Settings > Extensions > Cline > Custom Instructions

Paste the contents of SKILL.md into the text field.

Aider

No progressive loading — all files load at session start.

Option A: Convention File

# Copy to project root
cp path/to/progressive-estimation/SKILL.md CONVENTIONS.md

Add to .aider.conf.yml:

read: CONVENTIONS.md

Option B: Multiple Reference Files

# Copy the skill into your project
cp -r path/to/progressive-estimation ./estimation

Add to .aider.conf.yml:

read:
  - estimation/SKILL.md
  - estimation/references/formulas.md
  - estimation/references/questionnaire.md

Warning

All read: files load into context at session start and stay there. Only include files you'll actively use to preserve context window space. The full skill (~2,000 lines across all files) may consume significant context in smaller models.

Option C: On-Demand Loading

Start with just the workflow:

read: estimation/SKILL.md

Then load references as needed during the session:

/read estimation/references/formulas.md

Continue.dev

No progressive loading — rules go into every system message.

Add to ~/.continue/config.yaml:

rules:
  - uses: file:///absolute/path/to/progressive-estimation/SKILL.md

Or for a project-specific setup, create .continuerc.json:

{
  "rules": [
    {
      "uses": "file://./progressive-estimation/SKILL.md"
    }
  ]
}

Note

Rules are included in every request's system message. Keep it to SKILL.md only and instruct the model to read reference files from disk when needed during the estimation workflow.

ChatGPT (Projects)

No file-based config — everything is in the UI.

  1. Create a new Project in ChatGPT
  2. Open Project Instructions
  3. Paste the contents of SKILL.md into the instructions field
  4. Upload reference files to the Project:
    • references/formulas.md
    • references/questionnaire.md
    • references/frameworks.md
    • references/output-schema.md
    • references/calibration.md

Warning

Project instructions have a 1,500 character limit. You'll need a condensed version of SKILL.md for the instructions field. Upload the full SKILL.md as a project file instead, and use a short instruction like: "For estimation tasks, follow the workflow in the uploaded SKILL.md file and reference the uploaded formulas, questionnaire, and other files as needed."

Gemini Code Assist

No progressive loading — style guide loads with every request.

# Create the Gemini config directory
mkdir -p .gemini

# Copy as style guide
cp path/to/progressive-estimation/SKILL.md .gemini/styleguide.md

Note

Gemini Code Assist treats styleguide.md as a general coding style guide. For estimation-specific use, prepend a note: "This style guide includes an estimation workflow. Apply it when the user asks for estimates, sizing, or effort assessment."

For reference files, place them in the project and ask Gemini to read them during estimation conversations.

Consolidated Single-File Version

For clients with strict character limits or no multi-file support, you can ask Claude Code to generate a consolidated version:

Combine SKILL.md and the key formulas into a single file under 6000 characters

This produces a condensed version suitable for Windsurf, ChatGPT instructions, or any client with tight limits.

Comparison

Client Install Method Progressive Loading Limits Effort
Claude Code git clone to skills dir Full 2% context for descriptions Easiest
Cursor .cursor/rules/*.mdc Yes (agent-decided) No documented limit Easy
GitHub Copilot .github/instructions/ Yes (path-specific) No documented limit Easy
Windsurf .windsurf/rules/*.md Partial 6K/file, 12K total Moderate
Cline .clinerules/ directory No No documented limit Easy
Aider read: in .aider.conf.yml No (on-demand via /read) Context window Moderate
Continue.dev rules: in config.yaml No Context window Easy
ChatGPT Project UI + file uploads No 1,500 chars instructions Manual
Gemini .gemini/styleguide.md No No documented limit Easy