project-recon

Zero-dependency shell recon for any code repository — detect languages, count LOC, and report project scale. Pure POSIX find/wc or PowerShell, no Python or third-party tools required. Triggers: "how big is this project", "what languages", "project sizing", "repo recon", "LOC count", "scope check".

Project Recon: Zero-Dependency Repo Sizing

What this skill does

Quickly answers: how big is this repo and what's in it. Shell-only (POSIX find + wc or PowerShell), runs anywhere, no install needed.

Output Format

Emit exactly this JSON structure — no additional fields:

{
  "total_loc": <int>,
  "languages": { "<lang>": <loc>, ... },
  "top_level_dirs": <int>,
  "primary_language": "<lang with highest LOC>",
  "git": <bool>
}

total_loc: sum of non-blank lines across all detected source files
languages: per-language LOC breakdown (only languages actually found)
top_level_dirs: count of immediate subdirectories under repo root (excluding hidden dirs)
primary_language: the language key with the highest LOC
git: whether the project root is a git repository (git rev-parse --git-dir succeeds)

Do NOT add fields beyond this schema. No descriptions, no module lists, no domain analysis.

Files

File	Purpose
`references/language-extensions.yaml`	Language → file extension mapping + generated-file suffixes to skip
`references/exclude-patterns.yaml`	Directory exclude rules (global + per-language)
`references/loc-shell.md`	bash + PowerShell counting templates

Workflow

Detect languages. Scan for manifest files (pom.xml, *.csproj, package.json, go.mod, etc.) or dominant extensions. Look up language-extensions.yaml for the canonical extension list.
Resolve excludes. Union exclude-patterns.yaml::global.dirs with per_language.<lang>.dirs for each detected language.
Count. Use the template from references/loc-shell.md. Run once per detected language, sum for total.
Emit JSON. Output the schema above. Nothing else.

Semantics

Counts non-blank lines (wc -l), comment-inclusive. This is intentional — speed and simplicity over precision.
Expect 5–15% upward bias vs comment-stripping counters on verbose languages (Java, C#). Acceptable for sizing.