---
name: Project Discovery
slug: project-discovery-2
category: Automation
description: Project Discovery scans a codebase to identify its languages, frameworks, layout, and run commands, then writes a short reference into AGENTS.md or CLAUDE.md. Use it when you need a concise map of where things live and how to work in the project.
github: "https://github.com/testdouble/han/tree/main/han-core/skills/project-discovery"
language: Shell
stars: 253
forks: 22
install: "npx degit https://github.com/testdouble/han/tree/main/han-core/skills/project-discovery ~/.claude/skills/project-discovery"
installs_to: ~/.claude/skills/project-discovery
source_path: han-core/skills/project-discovery/SKILL.md
collection_size: 25
category_size: 1648
collection_url: "https://dirskills.com/collections/testdouble/han"
added: 2026-09-02T05:20:53.236Z
last_synced: 2026-09-02T05:20:53.236Z
canonical_url: "https://dirskills.com/skills/project-discovery-2"
---

# Project Discovery

Project Discovery scans a codebase to identify its languages, frameworks, layout, and run commands, then writes a short reference into AGENTS.md or CLAUDE.md. Use it when you need a concise map of where things live and how to work in the project.

**Install:**

```bash
npx degit https://github.com/testdouble/han/tree/main/han-core/skills/project-discovery ~/.claude/skills/project-discovery
```

## README

## Project Context

- Default branch: !`git symbolic-ref --short refs/remotes/origin/HEAD 2>/dev/null || echo unknown`
- AGENTS.md: !`find . -maxdepth 1 -name "AGENTS.md" -type f`
- CLAUDE.md: !`find . -maxdepth 1 -name "CLAUDE.md" -type f`
- README: !`find . -maxdepth 1 -name "README*" -type f`
- personal config directory: !`bash "${CLAUDE_PLUGIN_ROOT}/scripts/han-config-dir.sh" 2>/dev/null || echo "$HOME/.claude"`
- project .han/config.md: !`cat .han/config.md 2>/dev/null || echo ""`

As your first action, use the Read tool on `.han/config.md` inside the `personal config directory` path above. A read
that returns no file is no personal configuration: continue silently. When that file or the `project .han/config.md`
probe supplies content, apply it per [config-rule.md](../../references/config-rule.md), which governs precedence
between the two files, relative-path resolution, and what to do with a file that reads but cannot be used.

# Project Discovery

This skill discovers the project's core attributes and writes them as a concise `## Project Discovery` section
**directly into the project's AGENTS.md or CLAUDE.md** — not a separate file. The output is small by design: a few notes
on where things live, the languages and frameworks, and the commands to run, so an AI agent working in the repo can find
its way around. It is not an exhaustive inventory.

## Step 1: Choose and read the target file

Pick the single file this discovery is written into, by priority:

1. If **AGENTS.md** exists (the label above is non-empty), the target is AGENTS.md.
2. Otherwise, if **CLAUDE.md** exists, the target is CLAUDE.md.
3. If neither exists, the target is a new **CLAUDE.md** at the repository root, which you will create in Step 4.

If the target file already exists, read it in full and build the **deduplication baseline**: a list of everything the
file already documents that this skill would otherwise write — directory and folder locations, languages, frameworks,
package manager, build/test/lint/dev commands, and the docs, ADR, and coding-standards directories. Note whether the
file already has a `## Project Discovery` section.

If the target is a new CLAUDE.md, there is nothing to deduplicate against.

## Step 2: Discover repository structure

Launch a `han-core:project-scanner` agent to determine whether the repository contains one project or many, and what
each project's boundaries are. Wait for the agent to complete.

From the agent's results, build a project list. Each entry has a project name (directory name, or repository name for a
root-level project), root path, and dependency manifest path.

## Step 3: Explore project attributes

Launch 3 `han-core:project-scanner` agents in parallel, each with a different focus area. Include the project list from
Step 2 in each agent's prompt so they know which roots to explore. Keep each agent on the core facts an AI agent needs
to navigate and run the project — not an exhaustive catalog of every config file.

**Agent 1 — Languages and Frameworks:** For each project, read the dependency manifest to identify languages and version
constraints. Determine the package manager from the lock file type. From dependencies, identify the
structural/architectural frameworks that define how the project is built (web, frontend, test, ORM/database). Ignore
utility packages. Note runtime version requirements.

**Agent 2 — Commands:** For each project, find the task runner or build definition and extract the actual commands for
installing dependencies, running tests, linting, building, and the dev server. Only record commands that actually exist
— no guesses.

**Agent 3 — Layout:** Map where the important things live: the main source directory or directories, the test location,
and the documentation, ADR, and coding-standards directories — do not assume names like "docs". Capture only the handful
of directories someone needs to know to find their way around the repo; skip incidental files.

After all 3 agents complete, merge their findings, deduplicate across agents, and organize by project. Separate
repository-level items (default branch, docs, ADRs, coding standards, layout) from project-level items (language,
frameworks, package manager, commands).

## Step 4: Write the discovery into the target file

Build a concise `## Project Discovery` section using the template at [template.md](./references/template.md) as the
structural guide.

Apply two filters before writing anything:

- **Deduplicate.** Drop every fact already present in the target file (the Step 1 baseline). Do not restate what the
  file already says, even in different words.
- **Drop empties.** Omit any line with no discovered value. Never leave a `{placeholder}` behind, and never invent a
  command or path that was not discovered.

If a discovered fact **contradicts** what the file already states (for example, the file says `make test` but no
Makefile was discovered), use `AskUserQuestion` to surface the contradiction and ask which is correct — the existing
file or the filesystem discovery. Update the content based on the answer.

Then write the result into the target file:

- **Target exists, no `## Project Discovery` section:** append the section at the end of the file (or another sensible
  location).
- **Target exists, already has a `## Project Discovery` section:** replace that section's body with the new,
  deduplicated content. Do not leave the old content alongside the new.
- **Neither AGENTS.md nor CLAUDE.md exists:** create CLAUDE.md at the repository root containing the section.

If, after deduplication, nothing meaningful remains to add, do **not** write an empty section. Tell the user the target
file already covers the project's core attributes, and stop.

## Step 5: Keep the `.han/config.md` pointer honest

Han skills read project-local overrides from the project's `.han/config.md` when a project carries one (see
[../../references/config-rule.md](../../references/config-rule.md)). This step keeps the target file's pointer to that
config accurate, using the same consent gate and deduplication discipline as Step 4. The Project Context probe above
shows whether the file exists.

- **`.han/config.md` exists and the target file contains no reference to it:** use `AskUserQuestion` to offer adding a
  one-line pointer beside the `## Project Discovery` section, for example:
  `Han skills in this project read overrides from [.han/config.md](./.han/config.md).` Never add a pointer when any
  reference to the file is already present, even phrased differently.
- **`.han/config.md` does not exist but the target file still references it:** use `AskUserQuestion` to offer removing
  the stale pointer.
- **Otherwise:** do nothing and say nothing about the config.

Write or remove the pointer only with the user's consent.

## Step 6: Verification

Read back the target file's `## Project Discovery` section. Confirm: no `{placeholder}` text remains, every bullet has a
real discovered value, and nothing duplicates content stated elsewhere in the file. Spot-check 2-3 discovered paths or
directories with Glob to confirm they exist.

Report to the user: the target file written (and whether it was created or updated), the number of projects discovered,
and the languages and frameworks found.