# detect-ai-value-mode
An **agent skill** that reads a codebase and tells you *how its AI product
creates value* — and therefore how it should be positioned and sold. Point it at
your own repo or anyone else's, and it returns a sober, evidence-grounded verdict
report with every claim cited to a specific file or function.
It classifies the product against four value modes from the Enterprise AI
Adoption Framework:
| Mode | The product… | Pattern |
|------|--------------|---------|
| **Amplifier** | speeds up an existing task | copilot |
| **Substitute** | replaces a human function end to end | autonomous agent |
| **Job Creator** | enables a new professional role that did not exist | net-new |
| **Democratiser** | makes specialist work accessible to non-specialists | leveling |
This is a **skill for AI coding agents** (Claude Code, GitHub Copilot, Cursor,
Codex CLI, and anything that reads the [agentskills.io](https://agentskills.io)
open standard). It is not a standalone CLI — you install the skill into your
agent, then invoke it on a codebase.
> **No API key required.** The skill runs on the agent subscription you already
> have (Claude Code, Copilot, Codex, etc.). There is nothing to configure, no key
> to set, and no extra per-use cost — installing the folder is the whole setup.
---
## Install (about 20 seconds)
Copy the skill folder into your agent's skills directory. Keep the folder name
exactly `detect-ai-value-mode-from-codebase` (the skill name must match its folder).
```bash
git clone https://github.com/FrancyJGLisboa/detect-ai-value-mode.git
# Claude Code — personal (available in every project):
cp -R detect-ai-value-mode/detect-ai-value-mode-from-codebase ~/.claude/skills/
# …or project-scoped (just this repo):
cp -R detect-ai-value-mode/detect-ai-value-mode-from-codebase .claude/skills/
```
Other agents read the same folder from a different location — copy
`detect-ai-value-mode-from-codebase/` into whichever applies:
| Agent | Personal (all projects) | Project (one repo) |
|-------|-------------------------|--------------------|
| Claude Code | `~/.claude/skills/` | `.claude/skills/` |
| GitHub Copilot (VS Code & CLI) | `~/.copilot/skills/` *(Windows: `%APPDATA%\github-copilot\skills\`)* | `.github/skills/` |
| Codex CLI | `~/.codex/skills/` | `.codex/skills/` |
| Cursor | `~/.cursor/skills/` | `.cursor/skills/` |
Then restart or reload your agent so it picks up the new skill (in Copilot CLI:
`/skills reload`).
**One-command install for GitHub Copilot** (no clone, no copy):
```bash
gh skill install FrancyJGLisboa/detect-ai-value-mode detect-ai-value-mode-from-codebase
```
`gh skill install` (GitHub CLI, 2026-04+) drops the skill into the right directory
for your Copilot host automatically. Cursor and Codex also read each other's and
Claude's skill directories for compatibility, so a single `~/.claude/skills/` copy
is often picked up everywhere — but the table above is the canonical home per agent.
---
## Run it on a codebase
Inside your agent, invoke the skill with a target. Three ways to point it:
```text
# 1. Your own codebase — run from inside the project
cd ~/work/my-ai-product
/detect-ai-value-mode-from-codebase .
# 2. Any local path (your machine, doesn't need to be the current dir)
/detect-ai-value-mode-from-codebase ~/work/some-other-repo
# 3. Someone else's codebase by URL (it shallow-clones, reads, discards)
/detect-ai-value-mode-from-codebase https://github.com/org/their-repo
```
You can also just **describe the intent** in natural language and the agent will
trigger the skill from its description — e.g. *"what value mode is this codebase?"*
or *"read this repo and tell me how it creates value and who the buyer is."*
Pasting or uploading code works too.
### What you get (excerpt)
```text
VERDICT: SUBSTITUTE (primary) / AMPLIFIER (secondary) Confidence: MODERATE
This product replaces a human triage function rather than accelerating one.
The evidence is in agents/triage.py, where classify_and_route() acts on model
output with no human-in-the-loop checkpoint in the normal path; the approval
gate in review.py:guard() is exercised only on the low-confidence branch...
```
Full scoring rubric: [`references/value_modes.md`](detect-ai-value-mode-from-codebase/references/value_modes.md).
Exact report schema: [`references/verdict_schema.md`](detect-ai-value-mode-from-codebase/references/verdict_schema.md).
---
## Repository layout
```
detect-ai-value-mode/
├── detect-ai-value-mode-from-codebase/ ← the skill (this is what you install)
│ ├── SKILL.md ← instructions + metadata the agent reads
│ ├── evals.yaml ← trigger eval set (should-fire / should-not-fire)
│ └── references/ ← scoring rubric + report schema
└── tools/
└── validate_skill.py ← pre-publish validator + trigger-eval harness
```
## Validate before you change it
The skill ships green against the agentskills.io standard and the Anthropic
importer. If you fork or edit it, re-check with the bundled validator (zero
install — `uv` resolves its one dependency on first run):
```bash
uv run tools/validate_skill.py detect-ai-value-mode-from-codebase
```
Spec validation needs no key. To also check *trigger reliability*, ask your agent
to score `evals.yaml` against the description (uses your subscription, no key) —
see [`tools/README.md`](tools/README.md). An optional `--eval` API path exists only
for headless CI.
See [`tools/README.md`](tools/README.md) for the full gate list and CI wiring.
---
## Attribution & license
The four value modes are drawn from the Enterprise AI Adoption Framework
attributed to **Derley Morais (Iron Flow)**. This repository packages that
framework as a repeatable analysis workflow; the framework itself remains the
property of its author.
Skill packaging and tooling: **MIT** — see [`LICENSE`](LICENSE).