1bcoder 0.1.1__tar.gz → 0.1.3__tar.gz

{1bcoder-0.1.1 → 1bcoder-0.1.3/1bcoder.egg-info}/PKG-INFO

@@ -1,6 +1,9 @@
 Metadata-Version: 2.4
 Name: 1bcoder
-Version: 0.1.1
+Version: 0.1.3
+Summary: AI coding assistant agent for 1B–7B local models (Ollama, LMStudio, llama.cpp). Terminal REPL with file editing, project map, agents, scripts, and parallel multi-model queries.
+Project-URL: Homepage, https://github.com/szholobetsky/1bcoder
+Project-URL: Repository, https://github.com/szholobetsky/1bcoder
 Requires-Python: >=3.10
 Description-Content-Type: text/markdown
 License-File: LICENSE
@@ -12,15 +15,7 @@ Dynamic: license-file
 
 # 1bcoder
 
-AI-assisted code editor designed for small (1B parameter) language models running locally via [Ollama](https://ollama.com), [LMStudio](https://lmstudio.ai), or [LiteLLM](https://litellm.ai).
-
----
-
-**(c) 2026 Stanislav Zholobetskyi**
-Institute for Information Recording, National Academy of Sciences of Ukraine, Kyiv
-
-*Створено в рамках аспірантського дослідження на тему:
-«Інтелектуальна технологія підтримки розробки та супроводу програмних продуктів»*
+AI coding assistant agent for 1B–7B local models running locally via [Ollama](https://ollama.com), [LMStudio](https://lmstudio.ai), or [LiteLLM](https://litellm.ai).
 
 ---
 
@@ -93,11 +88,12 @@ Tasks that require the model to decide *what to look at* — refactoring across
 - **`/plan <goal>`** — planning agent: researches the project, writes a natural-language step-by-step plan to `plan.txt`; run `/agent <task> plan plan.txt` to execute it step by step
 - **`/fill`** — fill agent: reads NaN session variables, scans project for `.var` files and config files, sets each value automatically
 - **Session variables** — `{{name}}` placeholders substituted in any command; save/load from `.var` files for offline reuse without loading files into context
-- **Project config** — `/config save` persists session state (host, model, ctx, params, vars, procs) to `.1bcoder/config.yml`; auto-loaded on startup when `auto: true`
+- **Project config** — `/config save` persists session state (host, model, ctx, params, vars, procs) to `.1bcoder/config.yml`; `/config save global` saves to `~/.1bcoder/config.yml`; on startup, the first config with `auto: true` (local → global) is applied automatically
 - **Aliases** — define command shortcuts with `/alias /name = expansion` (supports `{{args}}`); persisted in `aliases.txt`; loaded from global then project directory at startup and survive `/clear`
 - **Backup/restore** — `/bkup save` rotates existing backups (`file.bkup` → `file.bkup(1)`, `file.bkup(2)`…) so no snapshot is ever overwritten; `/bkup restore` always restores the latest
 - **MCP support** — connect external tool servers (filesystem, web, git, database, browser…) via the Model Context Protocol
-- **Parallel queries** — send prompts to multiple models simultaneously with `/parallel`, with saved profiles
+- **Parallel queries** — send prompts to multiple models simultaneously with `/parallel`; control context sent (`--ctx`/`--last`/`--no-ctx`) and route replies back into main context (`ctx` output) for sub-agent workflows
+- **Command hooks** — `/hook before|after <cmd> <script>` runs a script before or after edit/patch/fix/insert; `before` hook cancels the command if the script is missing; `{{file}}` and `{{range}}` injected automatically
 - Switch model or host at runtime without restarting (`/model gemma3:1b`, `/host openai://localhost:1234`)
 - **Model parameters** — `/param temperature 0.2`, `/param enable_thinking false` — sent with every request, auto-cast to correct type
 - **Multi-provider** — connect to Ollama, LMStudio, or LiteLLM using `ollama://` / `openai://` URL scheme; plain host defaults to Ollama
@@ -112,7 +108,7 @@ Tasks that require the model to decide *what to look at* — refactoring across
 pip install 1bcoder
 ```
 
-On first launch, default agents, procs, and scripts are copied to `~/.1bcoder/` automatically.
+On first launch, default agents, procs, scripts, profiles, and aliases are copied to `~/.1bcoder/` automatically. On upgrade (`pip install --upgrade 1bcoder`), new entries in `aliases.txt` and `profiles.txt` are merged in without overwriting your customisations.
 
 ### Option 2 — Clone and install locally
 
@@ -184,15 +180,15 @@ pip install -e .
 python chat.py
 ```
 
-On startup a numbered list of available Ollama models is shown — type the number to select one. Use `--model` to skip the prompt.
+On startup, 1bcoder checks for a config with `auto: true` (local `.1bcoder/config.yml` first, then `~/.1bcoder/config.yml`) and connects to the host and model stored there. If no config is found, it connects to local Ollama and prompts for a model. Use `--model` or `--host` to override.
 
 ### CLI options
 
 ```
 1bcoder [--host URL] [--model NAME] [--init] [--scriptapply SCRIPT] [--param KEY=VALUE]
 
---host URL              Host URL — supports ollama:// and openai:// schemes (default: http://localhost:11434)
---model NAME            Skip model selection, use this model directly
+--host URL              Host URL — supports ollama:// and openai:// schemes (default: from config or http://localhost:11434)
+--model NAME            Model to use; overrides config (shows list if not available on host)
 --init                  Create .1bcoder/ scaffold in the current directory
 --scriptapply SCRIPT        Run a script file non-interactively, then exit
 --param KEY=VALUE       Plan parameter substitution (repeatable)
@@ -682,7 +678,7 @@ Lines starting with `[v]` are already done and skipped. Lines starting with `#`
 | `/script show` | Display steps of the current script |
 | `/script add <command>` | Append a step to the current script |
 | `/script clear` | Wipe current script completely |
-| `/script reset` | Unmark all done steps |
+| `/script reset` | Unmark all done steps (also happens automatically when a script runs to completion) |
 | `/script reapply [key=value ...]` | Reset all done steps then apply automatically; prompts for any NaN `{{variables}}` before running |
 | `/script refresh` | Reload script from disk and show contents |
 | `/script apply [file] [key=value ...]` | Run steps one by one (Y/n/q per step) |
@@ -732,6 +728,7 @@ Session variables store named values that are substituted as `{{name}}` in any c
 /var set name =MyService         literal value
 /var def port db host            declare multiple NaN variables (skips if already set)
 /var get                         list all variables (NaN = unset)
+/var get port                    print value of a single variable (useful with ->)
 /var del port                    remove a variable
 ```
 
@@ -777,9 +774,12 @@ Any `{{key}}` found but not yet set is registered as NaN — `/script reapply` w
 
 ---
 
-### Output capture (`->` and `$`)
+### Output capture (`->`, `$` and `~`)
+
+Any command — LLM reply, tool output, or proc result — can be captured into a session variable using the `->` suffix. Two special tokens expand anywhere in a command or message:
 
-Any command — LLM reply, tool output, or proc result — can be captured into a session variable using the `->` suffix. The special token `$` expands to the last captured output anywhere in a command or message.
+- `$` — last captured output (last AI reply or tool result)
+- `~` — last user input (last message or command you typed)
 
 ```
 /map keyword extract auth.py -> keywords      # capture tool output into variable
@@ -795,23 +795,43 @@ summarize this for me -> myplan              # capture LLM reply
 /var set port result                         # also works: grab key from proc output
 ```
 
+**`~` — repeat or redirect the last question:**
+```
+як працює цей метод?                         # ask main model
+/small ~                                     # same question → small model
+/ask ~                                       # same question → agent mode
+/explain "$"                                 # ask small model to explain the reply
+поясни: $                                    # ask main model to explain its own reply
+```
+
 `->` stores the full text (including ANSI-stripped terminal output) and also updates `$` for immediate reuse. Variables captured with `->` appear in `/var get` like any other session variable.
 
 ---
 
 ### Project config (`/config`)
 
-Save and restore session state (host, model, ctx, params, vars, procs) to `.1bcoder/config.yml` in the current working directory. Useful for project-specific presets that are too large to fit in model context.
+Save and restore session state (host, model, ctx, params, vars, procs). Two config locations are supported:
+
+- **Local** — `.1bcoder/config.yml` in the current working directory (project-specific)
+- **Global** — `~/.1bcoder/config.yml` (user-wide default for all projects)
+
+**Startup priority:** on launch without `--host`/`--model`, 1bcoder checks local config first, then global. The first one with `auto: true` wins. If neither has `auto: true`, connects to local Ollama and prompts for a model.
 
 ```
-/config save                     # save all current state
-/config save host                # save only host
-/config save model               # save only model
-/config save vars                # save only vars
-/config load                     # restore from config.yml
-/config show                     # print config.yml contents
-/config auto on                  # auto-load on every startup in this directory
-/config auto off                 # disable auto-load
+/config save                     # save all current state to local config
+/config save global              # save all current state to global config
+/config save host                # save only host to local config
+/config save global host         # save only host to global config
+/config save model               # save only model to local config
+/config save global model        # save only model to global config
+/config save vars                # save only vars to local config
+/config load                     # restore from local config
+/config load global              # restore from global config
+/config show                     # print local config contents
+/config show global              # print global config contents
+/config auto on                  # enable auto-load in local config
+/config auto on global           # enable auto-load in global config
+/config auto off                 # disable auto-load in local config
 ```
 
 **Selective delete:**
@@ -824,7 +844,7 @@ Save and restore session state (host, model, ctx, params, vars, procs) to `.1bco
 /config del proc collect-files   # remove one proc
 ```
 
-**Config file format** (`.1bcoder/config.yml`):
+**Config file format** (`.1bcoder/config.yml` or `~/.1bcoder/config.yml`):
 ```yaml
 auto: true
 host: ollama://localhost:11434
@@ -840,7 +860,7 @@ procs:
   - collect-files output.txt
 ```
 
-When `auto: true`, the config is applied automatically after the startup banner — host, model, ctx, params, vars, and procs are restored without any command.
+When `auto: true`, host and model are used at startup to connect; ctx, params, vars, and procs are also restored.
 
 ---
 
@@ -863,24 +883,34 @@ Connect external tool servers to give the AI access to filesystems, databases, w
 /mcp disconnect fs
 ```
 
-See [MCP.md](MCP.md) for a full list of ready-to-use servers.
+See `/doc MCP` for a full list of ready-to-use servers.
 
 ---
 
 ### Parallel queries
 
-Send prompts to multiple models at the same time. Each answer is saved to its own file.
+Send a prompt to multiple models at the same time.
 
 ```
-/parallel ["prompt"] [profile <name>] [host:port|model|file ...]
+/parallel ["prompt"] [--ctx|--last|--no-ctx] [profile <name>] [host:port|model|(file or ctx) ...]
 ```
 
+| Flag | Behaviour |
+|---|---|
+| *(default)* | Full conversation context is sent to every worker |
+| `--last` | Only the last user message is sent (saves tokens for small models) |
+| `--no-ctx` | No context — prompt only (fastest, zero leakage) |
+
+Workers write results to a file **or** inject them back into the main context:
+
 ```
 /parallel "review this for bugs" \
-    localhost:11434|llama3.2:1b|answers/llm1.txt \
-    localhost:11435|qwen2.5:1b|answers/llm2.txt
+    localhost:11434|llama3.2:1b|ans/llm1.txt \
+    localhost:11435|qwen2.5:1b|ctx
 ```
 
+Using `ctx` as the output target injects the worker's reply into the main conversation — the next AI turn will see it.
+
 **Profiles** — save a set of workers for reuse:
 
 ```
@@ -898,6 +928,60 @@ review: localhost:11434|ministral3:3b|ans/review.txt localhost:11435|cogito:3b|a
 fast:   localhost:11434|qwen2.5-coder:0.6b|ans/q.txt                                          # quick sanity check
 ```
 
+**Sub-agent profiles** — built-in profiles that return answers directly to the main context (`ctx`):
+
+```
+small:    localhost:11434|qwen3:0.6b|ctx
+explain:  localhost:11434|gemma3:1b|ctx
+thinking: localhost:11434|lfm2.5-thinking:1.2b|ctx
+short:    localhost:11434|llama3.2:1b|ctx
+```
+
+These are aliased as `/small`, `/explain`, `/thinking`, `/short` — use them like sub-agents:
+
+```
+/small "what does this function return?" --no-ctx   # ask tiny model, no context bleed
+/explain "$"                                        # ask gemma to explain last reply
+/small ~                                            # repeat last question to a small model
+```
+
+`~` expands to the last message you typed; `$` expands to the last AI reply — combine them to build sub-agent pipelines without copy-pasting.
+
+---
+
+### Hooks (`/hook`)
+
+Run a script automatically **before** or **after** a command. Useful for backups before edits, linting after patches, or any pre/post workflow step.
+
+```
+/hook before <cmd> <script>   # run script before every <cmd>
+/hook after  <cmd> <script>   # run script after  every <cmd>
+/hook list                    # show active hooks
+/hook clear <cmd>             # remove hooks for <cmd>
+/hook clear                   # remove all hooks
+```
+
+`<cmd>` is the command name without the slash: `edit`, `patch`, `fix`, `insert`, `run`.
+
+**Two script types:**
+- `.txt` — 1bcoder script (sequence of commands). `{{file}}` and `{{range}}` are injected as session variables.
+- `.py` — Python guard subprocess. Receives trigger content on `stdin`, outputs `BLOCK:`/`ALERT:`/`ACTION:` lines.
+
+**Auto-injected for `.txt` scripts:**
+
+| Variable | Value |
+|---|---|
+| `{{file}}` | file argument of the triggering command |
+| `{{range}}` | line range (if specified), e.g. `10-25` |
+
+**Examples:**
+```
+/hook before edit /bkup {{file}}              # backup before every edit (.txt script)
+/hook before run sql_readonly_guard.py        # block dangerous SQL (.py guard)
+```
+
+Missing `.txt` script cancels a `before` hook. `.py` guard cancels only if it prints `BLOCK:`. Step errors inside `.txt` scripts do not cancel the command.
+
 ---
 
 ### Prompt templates
@@ -926,9 +1010,9 @@ Run a Python script against the last LLM reply. Useful for extracting filenames,
 /proc new my-proc                  # create a new processor from template
 ```
 
-**Processor protocol:** `stdin` = last LLM reply · `stdout` = result · `key=value` lines = extracted params · `ACTION: /command` = confirmed and executed (run mode only) · exit 1 = failure.
+**Processor protocol:** `stdin` = last LLM reply · `stdout` = result · `key=value` lines = extracted params · `ACTION: /command` = confirmed and executed (run mode only) · `ALERT: message` = warning printed, continues · `BLOCK: reason` = cancels the triggering command (hook mode only) · exit 1 = failure.
 
-Built-in processors in `<install>/.1bcoder/proc/`:
+Built-in processors in `~/.1bcoder/proc/`:
 
 | Processor | Purpose | Best mode |
 |---|---|---|
@@ -937,8 +1021,20 @@ Built-in processors in `<install>/.1bcoder/proc/`:
 | `extract-list` | Convert first bullet/numbered list in reply to comma-separated line | one-shot |
 | `grounding-check` | Score identifiers against `map.txt`, warn if <50% | persistent |
 | `collect-files` | Accumulate filenames to `.1bcoder/collected-files.txt` | persistent |
-| `md` | Render last reply as formatted Markdown in terminal (`pip install rich`) | one-shot |
+| `md` | Render last reply as formatted Markdown in terminal | one-shot |
 | `mdx` | Render last reply as Markdown + LaTeX (KaTeX) + Mermaid diagrams in browser | one-shot |
+| `ctx_cut` | Auto `/ctx cut` when context exceeds threshold (default 90%) | persistent |
+| `rude_words` | Alert if reply contains profanity (`ua` arg adds Ukrainian list) | persistent |
+| `secret_check` | Alert if reply contains sensitive names (google, anthropic…) | persistent |
+| `sql_readonly_guard` | Alert (proc) or block (hook) on write SQL statements | both |
+
+**Guard usage examples:**
+```
+/proc on ctx_cut 80                          # auto cut at 80%
+/proc on rude_words ua                       # profanity check + Ukrainian
+/proc on secret_check client=acme            # + custom keyword
+/hook before run sql_readonly_guard.py       # block /run with DELETE/DROP/UPDATE
+```
 
 See `/doc PROC` for the full protocol, built-in processor reference, and guide to writing your own.
 
@@ -1187,3 +1283,9 @@ For human input, the corrected command is shown with `[fix?]` and you are asked
 | text-generation-webui | Linux / Win | `--api` flag | 5000 | `openai://` | oobabooga UI, needs `--api` flag to expose OpenAI endpoint |
 | TabbyAPI | Linux / Win | built-in | 5000 | `openai://` | Focused on exl2/GPTQ quantized models, low VRAM |
 | vLLM | Linux | built-in | 8000 | `openai://` | Production server, high throughput, requires significant VRAM |
+
+---
+
+**(c) 2026 Stanislav Zholobetskyi**
+Institute for Information Recording, National Academy of Sciences of Ukraine, Kyiv
+*PhD research: «Intelligent Technology for Software Development and Maintenance Support»*

{1bcoder-0.1.1 → 1bcoder-0.1.3}/1bcoder.egg-info/SOURCES.txt

@@ -25,6 +25,7 @@ _bcoder_data/doc/PARAM.md
 _bcoder_data/doc/PROC.md
 _bcoder_data/proc/add-save.py
 _bcoder_data/proc/collect-files.py
+_bcoder_data/proc/ctx_cut.py
 _bcoder_data/proc/extract-code.py
 _bcoder_data/proc/extract-files.py
 _bcoder_data/proc/extract-list.py
@@ -32,6 +33,9 @@ _bcoder_data/proc/grounding-check.py
 _bcoder_data/proc/md.py
 _bcoder_data/proc/mdx.py
 _bcoder_data/proc/regexp-extract.py
+_bcoder_data/proc/rude_words.py
+_bcoder_data/proc/secret_check.py
+_bcoder_data/proc/sql_readonly_guard.py
 _bcoder_data/prompts/analysis.txt
 _bcoder_data/prompts/sumarise.txt
 _bcoder_data/scripts/AddFunction.txt

1bcoder-0.1.1/README.md → 1bcoder-0.1.3/PKG-INFO

@@ -1,14 +1,21 @@
-# 1bcoder
-
-AI-assisted code editor designed for small (1B parameter) language models running locally via [Ollama](https://ollama.com), [LMStudio](https://lmstudio.ai), or [LiteLLM](https://litellm.ai).
-
----
+Metadata-Version: 2.4
+Name: 1bcoder
+Version: 0.1.3
+Summary: AI coding assistant agent for 1B–7B local models (Ollama, LMStudio, llama.cpp). Terminal REPL with file editing, project map, agents, scripts, and parallel multi-model queries.
+Project-URL: Homepage, https://github.com/szholobetsky/1bcoder
+Project-URL: Repository, https://github.com/szholobetsky/1bcoder
+Requires-Python: >=3.10
+Description-Content-Type: text/markdown
+License-File: LICENSE
+Requires-Dist: requests>=2.28
+Requires-Dist: pyreadline3>=3.4; sys_platform == "win32"
+Requires-Dist: tqdm>=4.64
+Requires-Dist: rich>=13.0
+Dynamic: license-file
 
-**(c) 2026 Stanislav Zholobetskyi**
-Institute for Information Recording, National Academy of Sciences of Ukraine, Kyiv
+# 1bcoder
 
-*Створено в рамках аспірантського дослідження на тему:
-«Інтелектуальна технологія підтримки розробки та супроводу програмних продуктів»*
+AI coding assistant agent for 1B–7B local models running locally via [Ollama](https://ollama.com), [LMStudio](https://lmstudio.ai), or [LiteLLM](https://litellm.ai).
 
 ---
 
@@ -81,11 +88,12 @@ Tasks that require the model to decide *what to look at* — refactoring across
 - **`/plan <goal>`** — planning agent: researches the project, writes a natural-language step-by-step plan to `plan.txt`; run `/agent <task> plan plan.txt` to execute it step by step
 - **`/fill`** — fill agent: reads NaN session variables, scans project for `.var` files and config files, sets each value automatically
 - **Session variables** — `{{name}}` placeholders substituted in any command; save/load from `.var` files for offline reuse without loading files into context
-- **Project config** — `/config save` persists session state (host, model, ctx, params, vars, procs) to `.1bcoder/config.yml`; auto-loaded on startup when `auto: true`
+- **Project config** — `/config save` persists session state (host, model, ctx, params, vars, procs) to `.1bcoder/config.yml`; `/config save global` saves to `~/.1bcoder/config.yml`; on startup, the first config with `auto: true` (local → global) is applied automatically
 - **Aliases** — define command shortcuts with `/alias /name = expansion` (supports `{{args}}`); persisted in `aliases.txt`; loaded from global then project directory at startup and survive `/clear`
 - **Backup/restore** — `/bkup save` rotates existing backups (`file.bkup` → `file.bkup(1)`, `file.bkup(2)`…) so no snapshot is ever overwritten; `/bkup restore` always restores the latest
 - **MCP support** — connect external tool servers (filesystem, web, git, database, browser…) via the Model Context Protocol
-- **Parallel queries** — send prompts to multiple models simultaneously with `/parallel`, with saved profiles
+- **Parallel queries** — send prompts to multiple models simultaneously with `/parallel`; control context sent (`--ctx`/`--last`/`--no-ctx`) and route replies back into main context (`ctx` output) for sub-agent workflows
+- **Command hooks** — `/hook before|after <cmd> <script>` runs a script before or after edit/patch/fix/insert; `before` hook cancels the command if the script is missing; `{{file}}` and `{{range}}` injected automatically
 - Switch model or host at runtime without restarting (`/model gemma3:1b`, `/host openai://localhost:1234`)
 - **Model parameters** — `/param temperature 0.2`, `/param enable_thinking false` — sent with every request, auto-cast to correct type
 - **Multi-provider** — connect to Ollama, LMStudio, or LiteLLM using `ollama://` / `openai://` URL scheme; plain host defaults to Ollama
@@ -100,7 +108,7 @@ Tasks that require the model to decide *what to look at* — refactoring across
 pip install 1bcoder
 ```
 
-On first launch, default agents, procs, and scripts are copied to `~/.1bcoder/` automatically.
+On first launch, default agents, procs, scripts, profiles, and aliases are copied to `~/.1bcoder/` automatically. On upgrade (`pip install --upgrade 1bcoder`), new entries in `aliases.txt` and `profiles.txt` are merged in without overwriting your customisations.
 
 ### Option 2 — Clone and install locally
 
@@ -172,15 +180,15 @@ pip install -e .
 python chat.py
 ```
 
-On startup a numbered list of available Ollama models is shown — type the number to select one. Use `--model` to skip the prompt.
+On startup, 1bcoder checks for a config with `auto: true` (local `.1bcoder/config.yml` first, then `~/.1bcoder/config.yml`) and connects to the host and model stored there. If no config is found, it connects to local Ollama and prompts for a model. Use `--model` or `--host` to override.
 
 ### CLI options
 
 ```
 1bcoder [--host URL] [--model NAME] [--init] [--scriptapply SCRIPT] [--param KEY=VALUE]
 
---host URL              Host URL — supports ollama:// and openai:// schemes (default: http://localhost:11434)
---model NAME            Skip model selection, use this model directly
+--host URL              Host URL — supports ollama:// and openai:// schemes (default: from config or http://localhost:11434)
+--model NAME            Model to use; overrides config (shows list if not available on host)
 --init                  Create .1bcoder/ scaffold in the current directory
 --scriptapply SCRIPT        Run a script file non-interactively, then exit
 --param KEY=VALUE       Plan parameter substitution (repeatable)
@@ -670,7 +678,7 @@ Lines starting with `[v]` are already done and skipped. Lines starting with `#`
 | `/script show` | Display steps of the current script |
 | `/script add <command>` | Append a step to the current script |
 | `/script clear` | Wipe current script completely |
-| `/script reset` | Unmark all done steps |
+| `/script reset` | Unmark all done steps (also happens automatically when a script runs to completion) |
 | `/script reapply [key=value ...]` | Reset all done steps then apply automatically; prompts for any NaN `{{variables}}` before running |
 | `/script refresh` | Reload script from disk and show contents |
 | `/script apply [file] [key=value ...]` | Run steps one by one (Y/n/q per step) |
@@ -720,6 +728,7 @@ Session variables store named values that are substituted as `{{name}}` in any c
 /var set name =MyService         literal value
 /var def port db host            declare multiple NaN variables (skips if already set)
 /var get                         list all variables (NaN = unset)
+/var get port                    print value of a single variable (useful with ->)
 /var del port                    remove a variable
 ```
 
@@ -765,9 +774,12 @@ Any `{{key}}` found but not yet set is registered as NaN — `/script reapply` w
 
 ---
 
-### Output capture (`->` and `$`)
+### Output capture (`->`, `$` and `~`)
 
-Any command — LLM reply, tool output, or proc result — can be captured into a session variable using the `->` suffix. The special token `$` expands to the last captured output anywhere in a command or message.
+Any command — LLM reply, tool output, or proc result — can be captured into a session variable using the `->` suffix. Two special tokens expand anywhere in a command or message:
+
+- `$` — last captured output (last AI reply or tool result)
+- `~` — last user input (last message or command you typed)
 
 ```
 /map keyword extract auth.py -> keywords      # capture tool output into variable
@@ -783,23 +795,43 @@ summarize this for me -> myplan              # capture LLM reply
 /var set port result                         # also works: grab key from proc output
 ```
 
+**`~` — repeat or redirect the last question:**
+```
+як працює цей метод?                         # ask main model
+/small ~                                     # same question → small model
+/ask ~                                       # same question → agent mode
+/explain "$"                                 # ask small model to explain the reply
+поясни: $                                    # ask main model to explain its own reply
+```
+
 `->` stores the full text (including ANSI-stripped terminal output) and also updates `$` for immediate reuse. Variables captured with `->` appear in `/var get` like any other session variable.
 
 ---
 
 ### Project config (`/config`)
 
-Save and restore session state (host, model, ctx, params, vars, procs) to `.1bcoder/config.yml` in the current working directory. Useful for project-specific presets that are too large to fit in model context.
+Save and restore session state (host, model, ctx, params, vars, procs). Two config locations are supported:
+
+- **Local** — `.1bcoder/config.yml` in the current working directory (project-specific)
+- **Global** — `~/.1bcoder/config.yml` (user-wide default for all projects)
+
+**Startup priority:** on launch without `--host`/`--model`, 1bcoder checks local config first, then global. The first one with `auto: true` wins. If neither has `auto: true`, connects to local Ollama and prompts for a model.
 
 ```
-/config save                     # save all current state
-/config save host                # save only host
-/config save model               # save only model
-/config save vars                # save only vars
-/config load                     # restore from config.yml
-/config show                     # print config.yml contents
-/config auto on                  # auto-load on every startup in this directory
-/config auto off                 # disable auto-load
+/config save                     # save all current state to local config
+/config save global              # save all current state to global config
+/config save host                # save only host to local config
+/config save global host         # save only host to global config
+/config save model               # save only model to local config
+/config save global model        # save only model to global config
+/config save vars                # save only vars to local config
+/config load                     # restore from local config
+/config load global              # restore from global config
+/config show                     # print local config contents
+/config show global              # print global config contents
+/config auto on                  # enable auto-load in local config
+/config auto on global           # enable auto-load in global config
+/config auto off                 # disable auto-load in local config
 ```
 
 **Selective delete:**
@@ -812,7 +844,7 @@ Save and restore session state (host, model, ctx, params, vars, procs) to `.1bco
 /config del proc collect-files   # remove one proc
 ```
 
-**Config file format** (`.1bcoder/config.yml`):
+**Config file format** (`.1bcoder/config.yml` or `~/.1bcoder/config.yml`):
 ```yaml
 auto: true
 host: ollama://localhost:11434
@@ -828,7 +860,7 @@ procs:
   - collect-files output.txt
 ```
 
-When `auto: true`, the config is applied automatically after the startup banner — host, model, ctx, params, vars, and procs are restored without any command.
+When `auto: true`, host and model are used at startup to connect; ctx, params, vars, and procs are also restored.
 
 ---
 
@@ -851,24 +883,34 @@ Connect external tool servers to give the AI access to filesystems, databases, w
 /mcp disconnect fs
 ```
 
-See [MCP.md](MCP.md) for a full list of ready-to-use servers.
+See `/doc MCP` for a full list of ready-to-use servers.
 
 ---
 
 ### Parallel queries
 
-Send prompts to multiple models at the same time. Each answer is saved to its own file.
+Send a prompt to multiple models at the same time.
 
 ```
-/parallel ["prompt"] [profile <name>] [host:port|model|file ...]
+/parallel ["prompt"] [--ctx|--last|--no-ctx] [profile <name>] [host:port|model|(file or ctx) ...]
 ```
 
+| Flag | Behaviour |
+|---|---|
+| *(default)* | Full conversation context is sent to every worker |
+| `--last` | Only the last user message is sent (saves tokens for small models) |
+| `--no-ctx` | No context — prompt only (fastest, zero leakage) |
+
+Workers write results to a file **or** inject them back into the main context:
+
 ```
 /parallel "review this for bugs" \
-    localhost:11434|llama3.2:1b|answers/llm1.txt \
-    localhost:11435|qwen2.5:1b|answers/llm2.txt
+    localhost:11434|llama3.2:1b|ans/llm1.txt \
+    localhost:11435|qwen2.5:1b|ctx
 ```
 
+Using `ctx` as the output target injects the worker's reply into the main conversation — the next AI turn will see it.
+
 **Profiles** — save a set of workers for reuse:
 
 ```
@@ -886,6 +928,60 @@ review: localhost:11434|ministral3:3b|ans/review.txt localhost:11435|cogito:3b|a
 fast:   localhost:11434|qwen2.5-coder:0.6b|ans/q.txt                                          # quick sanity check
 ```
 
+**Sub-agent profiles** — built-in profiles that return answers directly to the main context (`ctx`):
+
+```
+small:    localhost:11434|qwen3:0.6b|ctx
+explain:  localhost:11434|gemma3:1b|ctx
+thinking: localhost:11434|lfm2.5-thinking:1.2b|ctx
+short:    localhost:11434|llama3.2:1b|ctx
+```
+
+These are aliased as `/small`, `/explain`, `/thinking`, `/short` — use them like sub-agents:
+
+```
+/small "what does this function return?" --no-ctx   # ask tiny model, no context bleed
+/explain "$"                                        # ask gemma to explain last reply
+/small ~                                            # repeat last question to a small model
+```
+
+`~` expands to the last message you typed; `$` expands to the last AI reply — combine them to build sub-agent pipelines without copy-pasting.
+
+---
+
+### Hooks (`/hook`)
+
+Run a script automatically **before** or **after** a command. Useful for backups before edits, linting after patches, or any pre/post workflow step.
+
+```
+/hook before <cmd> <script>   # run script before every <cmd>
+/hook after  <cmd> <script>   # run script after  every <cmd>
+/hook list                    # show active hooks
+/hook clear <cmd>             # remove hooks for <cmd>
+/hook clear                   # remove all hooks
+```
+
+`<cmd>` is the command name without the slash: `edit`, `patch`, `fix`, `insert`, `run`.
+
+**Two script types:**
+- `.txt` — 1bcoder script (sequence of commands). `{{file}}` and `{{range}}` are injected as session variables.
+- `.py` — Python guard subprocess. Receives trigger content on `stdin`, outputs `BLOCK:`/`ALERT:`/`ACTION:` lines.
+
+**Auto-injected for `.txt` scripts:**
+
+| Variable | Value |
+|---|---|
+| `{{file}}` | file argument of the triggering command |
+| `{{range}}` | line range (if specified), e.g. `10-25` |
+
+**Examples:**
+```
+/hook before edit /bkup {{file}}              # backup before every edit (.txt script)
+/hook before run sql_readonly_guard.py        # block dangerous SQL (.py guard)
+```
+
+Missing `.txt` script cancels a `before` hook. `.py` guard cancels only if it prints `BLOCK:`. Step errors inside `.txt` scripts do not cancel the command.
+
 ---
 
 ### Prompt templates
@@ -914,9 +1010,9 @@ Run a Python script against the last LLM reply. Useful for extracting filenames,
 /proc new my-proc                  # create a new processor from template
 ```
 
-**Processor protocol:** `stdin` = last LLM reply · `stdout` = result · `key=value` lines = extracted params · `ACTION: /command` = confirmed and executed (run mode only) · exit 1 = failure.
+**Processor protocol:** `stdin` = last LLM reply · `stdout` = result · `key=value` lines = extracted params · `ACTION: /command` = confirmed and executed (run mode only) · `ALERT: message` = warning printed, continues · `BLOCK: reason` = cancels the triggering command (hook mode only) · exit 1 = failure.
 
-Built-in processors in `<install>/.1bcoder/proc/`:
+Built-in processors in `~/.1bcoder/proc/`:
 
 | Processor | Purpose | Best mode |
 |---|---|---|
@@ -925,8 +1021,20 @@ Built-in processors in `<install>/.1bcoder/proc/`:
 | `extract-list` | Convert first bullet/numbered list in reply to comma-separated line | one-shot |
 | `grounding-check` | Score identifiers against `map.txt`, warn if <50% | persistent |
 | `collect-files` | Accumulate filenames to `.1bcoder/collected-files.txt` | persistent |
-| `md` | Render last reply as formatted Markdown in terminal (`pip install rich`) | one-shot |
+| `md` | Render last reply as formatted Markdown in terminal | one-shot |
 | `mdx` | Render last reply as Markdown + LaTeX (KaTeX) + Mermaid diagrams in browser | one-shot |
+| `ctx_cut` | Auto `/ctx cut` when context exceeds threshold (default 90%) | persistent |
+| `rude_words` | Alert if reply contains profanity (`ua` arg adds Ukrainian list) | persistent |
+| `secret_check` | Alert if reply contains sensitive names (google, anthropic…) | persistent |
+| `sql_readonly_guard` | Alert (proc) or block (hook) on write SQL statements | both |
+
+**Guard usage examples:**
+```
+/proc on ctx_cut 80                          # auto cut at 80%
+/proc on rude_words ua                       # profanity check + Ukrainian
+/proc on secret_check client=acme            # + custom keyword
+/hook before run sql_readonly_guard.py       # block /run with DELETE/DROP/UPDATE
+```
 
 See `/doc PROC` for the full protocol, built-in processor reference, and guide to writing your own.
 
@@ -1175,3 +1283,9 @@ For human input, the corrected command is shown with `[fix?]` and you are asked
 | text-generation-webui | Linux / Win | `--api` flag | 5000 | `openai://` | oobabooga UI, needs `--api` flag to expose OpenAI endpoint |
 | TabbyAPI | Linux / Win | built-in | 5000 | `openai://` | Focused on exl2/GPTQ quantized models, low VRAM |
 | vLLM | Linux | built-in | 8000 | `openai://` | Production server, high throughput, requires significant VRAM |
+
+---
+
+**(c) 2026 Stanislav Zholobetskyi**
+Institute for Information Recording, National Academy of Sciences of Ukraine, Kyiv
+*PhD research: «Intelligent Technology for Software Development and Maintenance Support»*