by the numbers

By the numbers

Snapshot of size, shape, and activity for this repository. Data collected 2026-08-06. Counts are approximate (whitespace and comments included; node_modules, dist, and out excluded). See Architecture for how the pieces fit together, and Overview for product context.

This repo holds two co-located trees: Foundry (apps/desktop/, the active product) and the SSSF skill (.claude/skills/sssf/, reference only).

Size by language

Approximate lines of code (LOC) for the two main bodies of work:

Language	Where	~LOC	Role
TypeScript / TSX	apps/desktop/ (src + tests)	~13,500	Foundry app and vitest suite
TypeScript (src only)	apps/desktop/src/	~11,700	Main, preload, renderer, shared
TypeScript (tests)	apps/desktop/tests/	~1,500	Seams against real git + fake-droid
CSS	apps/desktop/src/renderer/design/	~375	Design tokens
Python	.claude/skills/sssf/ (templates, scripts, modules)	~3,600	SSSF ADW skill (not linked by the app)
Vue / TS visualizer	.claude/skills/sssf/apps/visualizer/	templates + small SPA	Polled trace UI for stamped repos

xychart-beta horizontal
    title "Approx LOC by language (2026-08-06)"
    x-axis ["TS/TSX Foundry", "Python SSSF", "CSS Foundry"]
    y-axis "Lines" 0 --> 15000
    bar [13500, 3600, 375]

Loading

Config and packaging under apps/desktop/ (package.json, electron-vite, electron-builder, tsconfig, vitest) are small relative to source and are not charted.

Source vs tests vs config (Foundry)

Bucket	Path	~LOC	Notes
Application source	apps/desktop/src/	~11,700	Engine, droid, trace, store, system, renderer, shared, preload
Tests	apps/desktop/tests/	~1,500	7 files, ~94 tests claimed by AGENTS.md
Scripts	apps/desktop/scripts/	small	Headless engine:demo, icon helper
Config / packaging	root of apps/desktop/	small	Vite, Electron builder, TypeScript, Vitest

Rough ratio: source is about 8× the test suite by lines. Tests punch above that weight: they exercise real mkdtemp git repos and a scripted droid peer, not mocks of the engine loop.

TODOs in apps/desktop/src: none found on this date.

LOC by subsystem (Foundry src only)

Subsystem	Path	~LOC
Renderer	src/renderer/	3,737
Engine	src/main/engine/	2,338
Droid harness	src/main/droid/	1,774
Store	src/main/store/	1,120
Trace	src/main/trace/	993
Shared contract	src/shared/	559
System	src/main/system/	290
Preload	src/preload/	119
IPC + bootstrap	src/main/ipc.ts, main.ts, context.ts	remainder of main

The UI is the single largest subtree. The factory spine (engine + droid + trace) is comparable in aggregate and carries most of the behavioural complexity. Preload stays deliberately thin: named invoke only, no generic escape hatch.

xychart-beta horizontal
    title "Foundry src LOC by subsystem"
    x-axis ["Renderer", "Engine", "Droid", "Store", "Trace", "Shared", "System", "Preload"]
    y-axis "Lines" 0 --> 4000
    bar [3737, 2338, 1774, 1120, 993, 559, 290, 119]

Loading

Complexity hotspots

Largest TypeScript files in Foundry (line counts from the 2026-08-06 tree):

File	~Lines	Why it is large
src/main/engine/executor.ts	841	Full run loop: phases, retries, corrections, acceptance, finish()
src/main/trace/tracer.ts	825	Single writer for runs, phases, events, envelopes, gates, sessions
src/main/droid/client.ts	465	Long-lived stream-JSON-RPC session over stdio
src/main/ipc.ts	427	Entire renderer capability surface (invoke handlers)

Average size is uneven. Trace is two files that own the whole write path (~500 lines each on average). Engine spreads across nine modules; droid across seven. Renderer spreads across many small screens and components under components/ and screens/, so directory averages there understate any one file.

Directory	~LOC	Rough file count	Rough avg LOC / file
src/main/engine/	2,338	9	~260
src/main/droid/	1,774	7	~250
src/main/store/	1,120	7	~160
src/main/trace/	993	2	~500
src/main/system/	290	3	~100
src/shared/	559	2	~280
src/preload/	119	1	119

Tests

Suite file	Focus (from AGENTS.md)	~Count
envelopes.test.ts	zod seams, extraction, correction messages	12
boundary.test.ts	glob matching, three-state allow, revert	12
ipc-clone.test.ts	structured-clone bridge payloads	6
gates.test.ts	each gate's evidence, unknown-gate failure	19
droid-client.test.ts	wire protocol against fake-droid	24
executor.test.ts	run loop against real git temp repos	21
fake-droid.ts	scripted stdio peer (fixture, not a suite)	—

Total: ~94 tests, no network and no live model in the loop.

Built-in product surface

Kind	Count	Notes
Builtin agents	5	planner, builder, scout, reviewer, documenter (no tester)
Builtin pipelines	7	from single-agent prompt through full-sdlc
Gates	6	artifacts, non-empty files, JSON, diff claims, verdict, command
Envelope kinds	several	generic, plan, build, scout, review, document (+ custom fields)

Git activity

Default branch: main. The history is young and sparse on this date:

Date	Commit (summary)	What landed
2026-08-02	Initial SSSF skill tree	Python ADWs, cookbooks, Vue visualizer templates
2026-08-06	Foundry addition	Full Electron app under apps/desktop/, PLAN.md, app-side tests

Bot-attributed commits: effectively 0% of the recorded history (no dependabot/renovate-style commit series observed in this snapshot).

Volume is not a growth curve yet: two product-level landings a few days apart, rather than a long series of incremental merges. Treat LOC above as a cold-start footprint, not a multi-year accumulation.

What the numbers imply

• Foundry is a vertical slice, not a monorepo of many packages: one Electron app, one test harness style, one IPC contract.
• Complexity concentrates in executor, tracer, and droid client, which matches the architecture story (code owns the loop; one SQLite writer; observed wire protocol).
• The skill under .claude/ is real code (~3.6k Python plus visualizer templates) but remains out of the runtime graph. Size there should not be mistaken for app debt.

For narrative history rather than tables, see Lore. For odd invariants that numbers cannot show, see Fun facts.