# Load Project Extension

Pi extension that starts a project-memory loading turn for the current working directory.

See `docs/spec/LOAD_PROJECT.md` for behavior and `docs/arch/EXTENSION_ARCHITECTURE.md` for extension boundaries.

## Commands

- `/load` — compatibility full-load command with duplicate-command conflict handling.
- `/dd:load` — stable namespaced full-load command.

Both commands produce full reporting. Free-form arguments are query text and never change the mode. Compact rendering remains an internal detail of the separate global automatic project-memory loader; there is no public compact slash command. Automatic tool results retain complete content for the agent while the Pi TUI shows three lines by default and expands or collapses through the configured `app.tools.expand` keybinding (`Ctrl+O` by default). Explicit and automatic loads expose `Help: dotdotgod --help` as optional discovery guidance without reporting CLI installation status or requiring Help execution.

## Documentation Map And Query

Without arguments, Load discovers shared Markdown below `docs/` and renders a prefix-compressed tree through directory depth 5. It does not run semantic query.

With arguments, Load:

1. runs `dotdotgod query <root> "<arguments>" --limit 30 --json` when available;
2. includes the best-ranked chunk from each of at most 30 distinct Markdown files;
3. renders the documentation tree through directory depth 3;
4. falls back to README routing and targeted reads if query or shell execution is unavailable.

At a depth boundary, the map reports exact recursive omitted-directory and Markdown-file counts. `load.documentationSummary.exclude` controls both map and query corpus exclusions; defaults exclude `docs/plan/**` and `docs/archive/**` bodies. Load still treats `docs/plan/README.md` and `docs/archive/README.md` as local routing maps when available.

Focused query can refresh ignored repository cache files under `.dotdotgod/vectors/` and the runtime's user-level model cache. It does not modify source, documentation, or project config.

## Baseline Routing

The loader checks and routes through these maintained entrypoints when available:

- `AGENTS.md`
- `CLAUDE.md`
- `CODEX.md`
- `README.md`
- `docs/README.md`
- `docs/spec/README.md`
- `docs/test/README.md`
- `docs/arch/README.md`
- `docs/plan/README.md`
- `docs/archive/README.md`

The agent follows local README indexes, reads spec/test/architecture bodies selectively, inspects relevant active plans, and uses the archive README as a history map without scanning archive bodies by default.

## Notes

- `/load` may conflict with another extension, so `/dd:load` is always registered as the stable alias.
- `load.pinnedPaths` and `load.pinnedBodies` remain config compatibility fields but do not affect Load output.
- Explicit full loads and internal automatic compact loads share discovery and query behavior; only report detail differs.