docs: remove stale references to instructions primitive and ad-hoc prompts

The instructions primitive and the -p ad-hoc prompt flag were removed
but several docs still referenced them. A new user following the README
would hit errors trying to use `ralph new instruction` or `ralph run -p`.

- README: remove instructions section, fix "four primitives" → "three",
  remove -p examples, add named ralph example
- index.md: remove ad-hoc prompt example
- changelog.md: document instructions and -p removal in 0.1.7 breaking
  changes, fix stale "four primitive types" reference
- codebase-map.md: remove stale "inline text" and "ad-hoc text" refs

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Author: 2 people

README.md

@@ -44,12 +44,6 @@ ralph run       # Starts the loop (Ctrl+C to stop)
 
 That's it. Two commands.
 
-Or skip setup entirely with an ad-hoc prompt:
-
-```bash
-ralph run -n 1 -p "Add type hints to all public functions in src/"
-```
-
 ### What `ralph init` creates
 
 **`ralph.toml`** — tells ralphify what command to run:
@@ -70,7 +64,7 @@ Reads the prompt, pipes it to the agent, waits for it to finish, then does it ag
 ```bash
 ralph run          # Run forever
 ralph run -n 10    # Run 10 iterations then stop
-ralph run -p "Fix the login bug"   # Ad-hoc prompt, no RALPH.md needed
+ralph run docs     # Use a named ralph from .ralphify/ralphs/
 ```
 
 ### What it looks like
@@ -114,7 +108,7 @@ Read the full writeup: [Ralph Wiggum as a "software engineer"](https://ghuntley.
 
 ## Beyond the basic loop
 
-The simple loop works, but ralphify's real power comes from four primitives that live in the `.ralphify/` directory.
+The simple loop works, but ralphify's real power comes from three primitives that live in the `.ralphify/` directory.
 
 ### Checks — the self-healing loop
 
@@ -163,16 +157,6 @@ command: git log --oneline -10
 
 The command runs before each iteration. Use `{{ contexts.git-log }}` in your `RALPH.md` to control where the output appears.
 
-### Instructions — reusable rules
-
-Instructions are static text blocks (coding standards, commit conventions) you can toggle on and off without editing the prompt.
-
-```bash
-ralph new instruction code-style
-```
-
-Drop `{{ instructions }}` into `RALPH.md` to inject all enabled instructions.
-
 ### Ralphs — named task switcher
 
 Keep multiple ralphs for different jobs and switch between them at run time:

docs/changelog.md

@@ -14,6 +14,8 @@ Simplified the CLI, added a spinner during iterations, and removed the experimen
 
 - **Removed `ralph ui` subcommand and web dashboard** — the experimental web dashboard introduced in 0.1.6 has been removed. The dashboard depended on FastAPI, React, and SQLite, adding significant complexity for a feature that wasn't ready for production use. All dashboard-related API endpoints, persistent history, and WebSocket streaming have been removed. The core agent streaming (Claude Code auto-detection with `--output-format stream-json`) remains — it's used by the CLI for live activity tracking.
 - **Removed `ralph ralphs` subcommand** — `ralph new <name>` is now shorthand for `ralph new ralph <name>`.
+- **Removed instructions primitive** — the `instructions` primitive type introduced in 0.1.3 has been removed. Use contexts for injecting reusable rules into prompts instead. The `{{ instructions }}` and `{{ instructions.name }}` placeholders no longer resolve.
+- **Removed ad-hoc prompts (`-p` flag)** — the `--prompt` / `-p` flag on `ralph run` has been removed. The positional `[PROMPT]` argument now exclusively accepts a named ralph from `.ralphify/ralphs/`. To run a quick one-off task, create a named ralph with `ralph new`.
 
 ### Added
 
@@ -26,7 +28,7 @@ Simplified the CLI, added a spinner during iterations, and removed the experimen
 
 ### Improved
 
-- Introduced `Primitive` protocol in `_discovery.py` so all four primitive types (`Check`, `Context`, `Instruction`, `Ralph`) share a typed interface for discovery, filtering, merging, and display.
+- Introduced `Primitive` protocol in `_discovery.py` so all three primitive types (`Check`, `Context`, `Ralph`) share a typed interface for discovery, filtering, merging, and display.
 - Added generic `_discover_and_filter_enabled()` in the engine, replacing per-type boilerplate with a single code path bounded by the `Primitive` protocol.
 - Moved check result serialization into `CheckResult.to_event_data()` so event data formatting has a single source of truth.
 - Added missing event data fields (`result_text`, `detail`, `duration_formatted`) to API docs so library users see what the engine actually emits.

docs/contributing/codebase-map.md

@@ -53,13 +53,13 @@ ralph run
   │
   ├── cli.py:run() — parse options, print banner
   │   ├── Load config from ralph.toml
-  │   ├── Resolve prompt via ralphs.resolve_ralph_source() (name > file path > inline text > toml)
+  │   ├── Resolve prompt via ralphs.resolve_ralph_source() (name > file path > toml)
   │   └── Build RunConfig and call engine.run_loop()
   │
   └── engine.py:run_loop(config, state, emitter)
        ├── Discover checks, contexts from .ralphify/
        └── Loop:
-            ├── Read RALPH.md (or use ad-hoc text)
+            ├── Read RALPH.md
             ├── Run contexts → resolve {{ contexts.* }} placeholders
             ├── Append check failures from previous iteration (if any)
             ├── Pipe assembled prompt to agent command via subprocess
@@ -112,7 +112,7 @@ The CLI uses a `ConsoleEmitter` (defined in `_console_emitter.py`) that renders
 
 1. **`engine.py`** — The core run loop. Uses `RunConfig` and `RunState` (from `_run_types.py`) and `EventEmitter`. This is where iteration logic lives.
 2. **`_run_types.py`** — `RunConfig`, `RunState`, and `RunStatus`. These are the shared data types used by the engine, CLI, and manager. Separated so modules that only need the types don't pull in execution logic.
-3. **`cli.py`** — All CLI commands. Delegates to `engine.run_loop()` for the actual loop. Prompt source resolution (name vs. file path vs. inline) lives in `ralphs.py:resolve_ralph_source()`. Scaffold templates live in `_templates.py`. Terminal event rendering lives in `_console_emitter.py`.
+3. **`cli.py`** — All CLI commands. Delegates to `engine.run_loop()` for the actual loop. Prompt source resolution (name vs. file path) lives in `ralphs.py:resolve_ralph_source()`. Scaffold templates live in `_templates.py`. Terminal event rendering lives in `_console_emitter.py`.
 4. **`_frontmatter.py`** + **`_discovery.py`** — Frontmatter parsing and primitive discovery. `_frontmatter.py` handles YAML parsing and defines marker constants. `_discovery.py` defines the `Primitive` protocol, scans `.ralphify/` directories, and provides `merge_by_name()` for overlaying ralph-scoped primitives on globals. Understanding both is essential for working on checks/contexts/ralphs.
 5. **`resolver.py`** — Template placeholder logic used by contexts. Small file but critical.
 

docs/index.md

@@ -48,12 +48,6 @@ ralph run       # Starts the loop (Ctrl+C to stop)
 
 `ralph init` creates a config file and a starter prompt. `ralph run` reads the prompt, pipes it to the agent, waits for it to finish, and does it again. Edit `RALPH.md` while the loop is running — changes take effect on the next iteration.
 
-Or skip setup and pass a prompt directly:
-
-```bash
-ralph run "Add type hints to all public functions in src/" -n 1
-```
-
 ### What it looks like
 
 ```