gemcode 0.3.77__tar.gz → 0.3.80__tar.gz

{gemcode-0.3.77/src/gemcode.egg-info → gemcode-0.3.80}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: gemcode
-Version: 0.3.77
+Version: 0.3.80
 Summary: Local-first coding agent on Google Gemini + ADK
 Author: GemCode Contributors
 License:                                  Apache License
@@ -197,32 +197,82 @@ This document is the **authoritative reference** for CLI behavior, configuration
 
 ## Table of contents
 
-1. [Architecture](#architecture)
-2. [Requirements and install](#requirements-and-install)
-3. [First run](#first-run)
-4. [CLI commands](#cli-commands)
-5. [Main CLI flags](#main-cli-flags)
-6. [The `.gemcode/` directory](#the-gemcode-directory)
-7. [Project context: `GEMINI.md`](#project-context-geminimd)
-8. [Function tools (catalog)](#function-tools-catalog)
-9. [REPL: slash commands](#repl-slash-commands)
-10. [GemSkills](#gemskills)
-11. [Output styles and rules](#output-styles-and-rules)
-12. [Checkpoints, diff, and rewind](#checkpoints-diff-and-rewind)
-13. [Multi-root workspaces (`/add-dir`)](#multi-root-workspaces-add-dir)
-14. [Model routing and thinking](#model-routing-and-thinking)
-15. [Capabilities](#capabilities)
-16. [Permissions and interactive approval](#permissions-and-interactive-approval)
-17. [Hooks](#hooks)
-18. [Token budget, context, and compaction](#token-budget-context-and-compaction)
-19. [MCP](#mcp)
-20. [IDE bridge: `gemcode ide --stdio`](#ide-bridge-gemcode-ide-stdio)
-21. [Eval harness and autotune](#eval-harness-and-autotune)
-22. [Kaira scheduler](#kaira-scheduler)
-23. [Live audio](#live-audio)
-24. [Related components](#related-components)
-25. [Environment variables](#environment-variables)
-26. [Development and release](#development-and-release)
+1. [What GemCode is (detailed)](#what-gemcode-is-detailed)
+2. [Architecture](#architecture)
+3. [Requirements and install](#requirements-and-install)
+4. [First run](#first-run)
+5. [CLI commands](#cli-commands)
+6. [Main CLI flags](#main-cli-flags)
+7. [The `.gemcode/` directory](#the-gemcode-directory)
+8. [Project context: `GEMINI.md`](#project-context-geminimd)
+9. [Function tools (catalog)](#function-tools-catalog)
+10. [REPL: slash commands](#repl-slash-commands)
+11. [GemSkills](#gemskills)
+12. [Curated memory](#curated-memory)
+13. [Workspace trust](#workspace-trust)
+14. [Output styles and rules](#output-styles-and-rules)
+15. [Checkpoints, diff, and rewind](#checkpoints-diff-and-rewind)
+16. [Multi-root workspaces (`/add-dir`)](#multi-root-workspaces-add-dir)
+17. [Model routing and thinking](#model-routing-and-thinking)
+18. [Capabilities](#capabilities)
+19. [Permissions and interactive approval](#permissions-and-interactive-approval)
+20. [Hooks](#hooks)
+21. [Token budget, context, and compaction](#token-budget-context-and-compaction)
+22. [MCP](#mcp)
+23. [IDE bridge: `gemcode ide --stdio`](#ide-bridge-gemcode-ide-stdio)
+24. [Eval harness and autotune](#eval-harness-and-autotune)
+25. [Kaira scheduler](#kaira-scheduler)
+26. [Live audio](#live-audio)
+27. [Related components](#related-components)
+28. [Environment variables](#environment-variables)
+29. [Development and release](#development-and-release)
+
+---
+
+## What GemCode is (detailed)
+
+**GemCode** is a single Python package that exposes a **`gemcode`** CLI. Every invocation is anchored to a **project root** (`-C`, default current directory). From there it:
+
+- Builds a **GemCodeConfig** (model, capabilities, permissions, limits, paths).
+- Optionally loads **MCP** toolsets from `.gemcode/mcp.json`.
+- Constructs an ADK **Runner** with a root **LlmAgent** whose **system instruction** aggregates: global behavior, `GEMINI.md`, capability sections, tool manifest, optional **output style** and **rules**, the **skills manifest** (metadata only), and any **session-loaded GemSkills** (full bodies after `/gemskill`).
+- Persists conversation state in **SQLite** (`.gemcode/sessions.sqlite`) keyed by **`--session`** / REPL session id.
+
+### Modes of use
+
+| Mode | How | Best for |
+|------|-----|----------|
+| **One-shot CLI** | `gemcode -C repo "prompt"` | Scripts, CI-style runs, quick questions. |
+| **REPL** | `gemcode -C repo` (TTY, no prompt arg) | Long sessions, slash commands, exploration. |
+| **TUI** | Same as REPL with `GEMCODE_TUI=1` (default when supported) | Scrollback, styled output, slash **completion menu**. |
+| **IDE** | VS Code extension → `gemcode ide --stdio` | Editor-native Chat, diff apply, proposals. |
+| **Kaira** | `gemcode kaira` | Background queue of independent agent jobs. |
+| **Live audio** | `gemcode live-audio` | Voice → Gemini Live (separate from file-editing REPL). |
+
+### “Memory” in GemCode (three different things)
+
+| Mechanism | Where | Purpose |
+|-----------|--------|---------|
+| **Session history** | `sessions.sqlite` | Full turn-by-turn chat for a session id. |
+| **Curated memory** | `GEMCODE_MEMORY.md`, `GEMCODE_USER.md` (under `.gemcode/`) | Small, **human-approved** facts injected when memory features are on; distinct from noisy auto-memory. |
+| **Embedding memory** | `memories.jsonl` + ADK hooks (when enabled) | Retrieval-oriented storage when **`GEMCODE_ENABLE_MEMORY`** (and related) are on. |
+
+Use **`/curated`** in the REPL to preview the curated snapshot; tools such as **`read_curated_memory`** / **`remember_fact`** operate on that layer.
+
+### GemSkills at a glance (four ways)
+
+| Mechanism | Effect |
+|-----------|--------|
+| **`/create gemskill <name> [description]`** | Creates **`.gemcode/skills/<name>/SKILL.md`** scaffold (new skill on disk). |
+| **`/gemskill <name>`** | **Pins** the skill’s **full body** into the **system instruction** for the **current session** (until `/gemskill clear`, new session, or resume). Rebuilds the runner. |
+| **`/append gemskill <name> <request>`** | One model turn instructed to **edit** that skill file according to `<request>`. |
+| **`/skill <name>`**, **`/<name>`**, or tools **`load_skill`** | **One-shot** turn: inject expanded skill into the **user message** for that turn only (does not pin to system prompt). |
+
+The built-in **`batch`** skill is manual-only (`disable_model_invocation`); use **`/batch <goal>`** to run it.
+
+### Slash completion
+
+The REPL and TUI use a **canonical** command list for Tab completion (`repl_commands.SLASH_COMMANDS`). Some spellings (e.g. `/quit`, `/logs`, `/embed`) still work in **`process_repl_slash`** but may not appear as separate menu rows—see descriptions in **`/help`**.
 
 ---
 
@@ -280,6 +330,18 @@ Non-interactive environments (CI, pipes) must set `GOOGLE_API_KEY` explicitly an
 | `gemcode kaira` | Stdin-line → queued jobs scheduler (see [Kaira](#kaira-scheduler)). |
 | `gemcode ide --stdio` | **JSONL IDE protocol** on stdin/stdout for editor extensions (hidden entry; used by VS Code). |
 
+**Files with a prompt (multimodal):** attach one or more files Gemini can read (images, **PDF**, audio, video, plain text, etc.) for that message only:
+
+```bash
+gemcode -C . --attach ./screenshot.png "Why does this UI look wrong?"
+gemcode -C . --attach ./report.pdf "What are the main conclusions?"
+gemcode -C . --image a.png --image b.png "Compare these two layouts"
+```
+
+(`--image` is an alias of `--attach`.)
+
+In the **REPL / TUI**, queue files for the **next** message: `/attach path/to.pdf` (repeat for multiple), then type your question. Use `/attach list`, `/attach clear`. Aliases: `/image`, `/img`, `/file`.
+
 ---
 
 ## Main CLI flags
@@ -299,6 +361,7 @@ Non-interactive environments (CI, pipes) must set `GOOGLE_API_KEY` explicitly an
 | `--tool-combination-mode` | Gemini 3 **tool context circulation**: `deep_research\|always\|never\|auto`. |
 | `--mcp` | Load MCP toolsets from `.gemcode/mcp.json` (requires `pip install -e ".[mcp]"`). |
 | `--max-llm-calls` | Cap model↔tool iterations (`RunConfig.max_llm_calls`). |
+| `--attach PATH` | Attach file(s) for **this** CLI message only (repeat flag). Gemini-supported MIME (e.g. images, PDF, audio, video). Default max ~20 MiB each (`GEMCODE_MAX_ATTACHMENT_BYTES`). Alias: `--image`. |
 
 Kaira and `live-audio` accept overlapping options (project root, `--yes`, research/embeddings, etc.); run `gemcode kaira -h` / `gemcode live-audio -h` for full lists.
 
@@ -311,6 +374,7 @@ State is **project-local** (unless noted).
 | Path / artifact | Purpose |
 |-----------------|---------|
 | `sessions.sqlite` | ADK session service: conversation history for `--session` ids. |
+| `GEMCODE_MEMORY.md`, `GEMCODE_USER.md` | **Curated memory** (see [Curated memory](#curated-memory)). |
 | `audit.log` | JSONL audit: tool usage, model usage, terminal reasons, optional tool-use summaries. |
 | `tool-results/` | Offloaded large tool outputs; references like `tool_result:<sha256>`. |
 | `artifacts/` | File artifacts (ADK `FileArtifactService`). |
@@ -421,10 +485,16 @@ In interactive mode, lines starting with `/` are **slash commands** (see `repl_c
 
 | Command | Purpose |
 |---------|---------|
+| `/attach <path>`, `/attach list`, `/attach clear` | Queue file(s) for the **next** message (Gemini multimodal). Aliases: **`/image`**, **`/img`**, **`/file`**. |
+| `/trust`, `/trust on`, `/trust off` | Show, grant, or revoke **workspace trust** for the project root (stored in `~/.gemcode/trust.json`; required for file/shell/git tools). |
 | `/init` \| `/init force` | Analyze the repo and generate or overwrite `GEMINI.md`. |
 | `/cost` | Token usage and estimated cost for the session. |
 | `/notes`, `/notes clear`, `/notes edit` | View, clear, or edit `.gemcode/notes.md`. |
-| `/create gemskill <name> [description]` | Scaffold `.gemcode/skills/<name>/SKILL.md`. |
+| `/create gemskill <name> [description]` | **Create** a new GemSkill (scaffold `SKILL.md`). |
+| `/gemskill <name>` | **Load** an existing skill into the **session system prompt** (until `/gemskill clear` or new session). |
+| `/gemskill list` \| `/gemskill clear` | List skills or unload all session-loaded skills. |
+| `/append gemskill <name> <request>` | **Iterate** the skill file: model edits `SKILL.md` per your request. |
+| `/curated` | Show **curated memory** snapshot (`.gemcode/GEMCODE_MEMORY.md`, `GEMCODE_USER.md`). |
 | `/style`, `/style <name>\|off` | List or activate **output styles** (`.gemcode/output-styles/*.md`). |
 | `/rules` | Show **rule** files from `.gemcode/rules/` (with path gating). |
 | `/diff`, `/diff last`, `/diff cp_…` | Git diff, or **checkpoint → workspace** diff. |
@@ -437,7 +507,7 @@ In interactive mode, lines starting with `/` are **slash commands** (see `repl_c
 | Command | Purpose |
 |---------|---------|
 | `/help` | Short help. |
-| `/status` | Model, capabilities, thinking, limits, risk/context telemetry. |
+| `/status` | Model, capabilities, thinking, limits, risk/context telemetry; **`loaded_skills`** (session-pinned GemSkills from `/gemskill`). |
 | `/config` | Dump active config fields. |
 | `/session`, `/session list`, `/session name`, `/session resume`, `/session new` | Session management; `/clear` aliases `/session new`. |
 | `/compact`, `/compact <focus>` | Force context compaction / summarization. |
@@ -445,6 +515,12 @@ In interactive mode, lines starting with `/` are **slash commands** (see `repl_c
 | `/context` | Context pressure and token breakdown (includes styles, rules, skills manifest, touched paths). |
 | `/audit [N]` | Tail of `audit.log`. |
 | `/tools` | Tool inventory for current config. |
+| `/tools smoke` | Declaration compile check only (lists failures). |
+| `/eval`, `/eval llm` | **Eval harness**: tool smoke + `pytest` if a `tests/` tree exists; optional LLM golden prompts (costs tokens). Writes `.gemcode/evals/last_eval.json`. |
+| `/autotune init <tag>` | Create git branch `autotune/<tag>` (requires a git repo). |
+| `/autotune eval`, `… llm` | Run eval and append `.gemcode/evals/autotune_ledger.jsonl`. |
+| `/login` | How to run **`gemcode login`** (API key is set outside the REPL). |
+| `/live-audio` | How to run **`gemcode live-audio`** (mic → Gemini Live). |
 | `/doctor` | Environment sanity check. |
 | `/version` | Version string. |
 | `/exit` | Leave the REPL. |
@@ -462,6 +538,7 @@ In interactive mode, lines starting with `/` are **slash commands** (see `repl_c
 |---------|---------|
 | `/computer`, `/computer on\|off`, `/computer url` | Browser automation (Playwright). |
 | `/research`, `/research on\|off` | Deep research tools. |
+| `/maps`, `/maps on\|off` | Maps **grounding** toggle (runner rebuild). |
 | `/embeddings on\|off` | Semantic search tool. |
 | `/caps`, `/caps …` | View or bulk-toggle capabilities. |
 | `/memory`, `/memory on\|off` | Persistent memory. |
@@ -496,7 +573,33 @@ The TUI (when `GEMCODE_TUI=1` and terminal supports it) provides **slash complet
 - **Discovery:** Only **metadata** (name + description) is preloaded into context for token efficiency. Full body loads **on demand** via `/skill <name>`, `/<name>`, or tools `load_skill` / `list_skills`.
 - **Built-in:** **`batch`** — parallel large-change workflow (map → units → `run_subtask` → verify). Exposed as `/batch <goal>`; not auto-invoked by the model (`disable_model_invocation`).
 
-Use **`/create gemskill <name>`** to scaffold a new skill directory.
+- **`/create gemskill <name>`** — create a new skill directory.
+- **`/gemskill <name>`** — pin the full skill body into the system prompt for this session.
+- **`/append gemskill <name> <text>`** — one-shot turn for the model to revise that skill on disk.
+
+---
+
+## Curated memory
+
+**Curated memory** is a small, **intentional** text layer separate from raw session logs or embedding retrieval:
+
+- **Project facts:** `.gemcode/GEMCODE_MEMORY.md` (conventions, commands, architecture notes the team wants the agent to see).
+- **User preferences (per project):** `.gemcode/GEMCODE_USER.md`.
+
+Legacy filenames **`.gemcode/MEMORY.md`** and **`.gemcode/USER.md`** are still read if the new names are absent.
+
+Content is **sanitized** before append (length and sensitivity heuristics). The REPL command **`/curated`** prints a bounded snapshot of what would be injected. When the **memory** capability is enabled, curated material can be combined with broader memory tools—see **`remember_fact`**, **`read_curated_memory`** in the [function tools](#function-tools-catalog) table.
+
+---
+
+## Workspace trust
+
+File, shell, and related tools require the **project root** to be **trusted**. Trust is stored in **`~/.gemcode/trust.json`** (path configurable via **`GEMCODE_HOME`**), not inside the repo.
+
+- **`/trust`** — show whether the current root is trusted.
+- **`/trust on`** / **`/trust off`** — grant or revoke trust for **`cfg.project_root`**.
+
+Without trust, mutating and shell tools remain blocked even if **`--yes`** is set—this reduces accidental execution when the cwd is wrong (e.g. home directory).
 
 ---
 

{gemcode-0.3.77 → gemcode-0.3.80}/README.md

@@ -8,32 +8,82 @@ This document is the **authoritative reference** for CLI behavior, configuration
 
 ## Table of contents
 
-1. [Architecture](#architecture)
-2. [Requirements and install](#requirements-and-install)
-3. [First run](#first-run)
-4. [CLI commands](#cli-commands)
-5. [Main CLI flags](#main-cli-flags)
-6. [The `.gemcode/` directory](#the-gemcode-directory)
-7. [Project context: `GEMINI.md`](#project-context-geminimd)
-8. [Function tools (catalog)](#function-tools-catalog)
-9. [REPL: slash commands](#repl-slash-commands)
-10. [GemSkills](#gemskills)
-11. [Output styles and rules](#output-styles-and-rules)
-12. [Checkpoints, diff, and rewind](#checkpoints-diff-and-rewind)
-13. [Multi-root workspaces (`/add-dir`)](#multi-root-workspaces-add-dir)
-14. [Model routing and thinking](#model-routing-and-thinking)
-15. [Capabilities](#capabilities)
-16. [Permissions and interactive approval](#permissions-and-interactive-approval)
-17. [Hooks](#hooks)
-18. [Token budget, context, and compaction](#token-budget-context-and-compaction)
-19. [MCP](#mcp)
-20. [IDE bridge: `gemcode ide --stdio`](#ide-bridge-gemcode-ide-stdio)
-21. [Eval harness and autotune](#eval-harness-and-autotune)
-22. [Kaira scheduler](#kaira-scheduler)
-23. [Live audio](#live-audio)
-24. [Related components](#related-components)
-25. [Environment variables](#environment-variables)
-26. [Development and release](#development-and-release)
+1. [What GemCode is (detailed)](#what-gemcode-is-detailed)
+2. [Architecture](#architecture)
+3. [Requirements and install](#requirements-and-install)
+4. [First run](#first-run)
+5. [CLI commands](#cli-commands)
+6. [Main CLI flags](#main-cli-flags)
+7. [The `.gemcode/` directory](#the-gemcode-directory)
+8. [Project context: `GEMINI.md`](#project-context-geminimd)
+9. [Function tools (catalog)](#function-tools-catalog)
+10. [REPL: slash commands](#repl-slash-commands)
+11. [GemSkills](#gemskills)
+12. [Curated memory](#curated-memory)
+13. [Workspace trust](#workspace-trust)
+14. [Output styles and rules](#output-styles-and-rules)
+15. [Checkpoints, diff, and rewind](#checkpoints-diff-and-rewind)
+16. [Multi-root workspaces (`/add-dir`)](#multi-root-workspaces-add-dir)
+17. [Model routing and thinking](#model-routing-and-thinking)
+18. [Capabilities](#capabilities)
+19. [Permissions and interactive approval](#permissions-and-interactive-approval)
+20. [Hooks](#hooks)
+21. [Token budget, context, and compaction](#token-budget-context-and-compaction)
+22. [MCP](#mcp)
+23. [IDE bridge: `gemcode ide --stdio`](#ide-bridge-gemcode-ide-stdio)
+24. [Eval harness and autotune](#eval-harness-and-autotune)
+25. [Kaira scheduler](#kaira-scheduler)
+26. [Live audio](#live-audio)
+27. [Related components](#related-components)
+28. [Environment variables](#environment-variables)
+29. [Development and release](#development-and-release)
+
+---
+
+## What GemCode is (detailed)
+
+**GemCode** is a single Python package that exposes a **`gemcode`** CLI. Every invocation is anchored to a **project root** (`-C`, default current directory). From there it:
+
+- Builds a **GemCodeConfig** (model, capabilities, permissions, limits, paths).
+- Optionally loads **MCP** toolsets from `.gemcode/mcp.json`.
+- Constructs an ADK **Runner** with a root **LlmAgent** whose **system instruction** aggregates: global behavior, `GEMINI.md`, capability sections, tool manifest, optional **output style** and **rules**, the **skills manifest** (metadata only), and any **session-loaded GemSkills** (full bodies after `/gemskill`).
+- Persists conversation state in **SQLite** (`.gemcode/sessions.sqlite`) keyed by **`--session`** / REPL session id.
+
+### Modes of use
+
+| Mode | How | Best for |
+|------|-----|----------|
+| **One-shot CLI** | `gemcode -C repo "prompt"` | Scripts, CI-style runs, quick questions. |
+| **REPL** | `gemcode -C repo` (TTY, no prompt arg) | Long sessions, slash commands, exploration. |
+| **TUI** | Same as REPL with `GEMCODE_TUI=1` (default when supported) | Scrollback, styled output, slash **completion menu**. |
+| **IDE** | VS Code extension → `gemcode ide --stdio` | Editor-native Chat, diff apply, proposals. |
+| **Kaira** | `gemcode kaira` | Background queue of independent agent jobs. |
+| **Live audio** | `gemcode live-audio` | Voice → Gemini Live (separate from file-editing REPL). |
+
+### “Memory” in GemCode (three different things)
+
+| Mechanism | Where | Purpose |
+|-----------|--------|---------|
+| **Session history** | `sessions.sqlite` | Full turn-by-turn chat for a session id. |
+| **Curated memory** | `GEMCODE_MEMORY.md`, `GEMCODE_USER.md` (under `.gemcode/`) | Small, **human-approved** facts injected when memory features are on; distinct from noisy auto-memory. |
+| **Embedding memory** | `memories.jsonl` + ADK hooks (when enabled) | Retrieval-oriented storage when **`GEMCODE_ENABLE_MEMORY`** (and related) are on. |
+
+Use **`/curated`** in the REPL to preview the curated snapshot; tools such as **`read_curated_memory`** / **`remember_fact`** operate on that layer.
+
+### GemSkills at a glance (four ways)
+
+| Mechanism | Effect |
+|-----------|--------|
+| **`/create gemskill <name> [description]`** | Creates **`.gemcode/skills/<name>/SKILL.md`** scaffold (new skill on disk). |
+| **`/gemskill <name>`** | **Pins** the skill’s **full body** into the **system instruction** for the **current session** (until `/gemskill clear`, new session, or resume). Rebuilds the runner. |
+| **`/append gemskill <name> <request>`** | One model turn instructed to **edit** that skill file according to `<request>`. |
+| **`/skill <name>`**, **`/<name>`**, or tools **`load_skill`** | **One-shot** turn: inject expanded skill into the **user message** for that turn only (does not pin to system prompt). |
+
+The built-in **`batch`** skill is manual-only (`disable_model_invocation`); use **`/batch <goal>`** to run it.
+
+### Slash completion
+
+The REPL and TUI use a **canonical** command list for Tab completion (`repl_commands.SLASH_COMMANDS`). Some spellings (e.g. `/quit`, `/logs`, `/embed`) still work in **`process_repl_slash`** but may not appear as separate menu rows—see descriptions in **`/help`**.
 
 ---
 
@@ -91,6 +141,18 @@ Non-interactive environments (CI, pipes) must set `GOOGLE_API_KEY` explicitly an
 | `gemcode kaira` | Stdin-line → queued jobs scheduler (see [Kaira](#kaira-scheduler)). |
 | `gemcode ide --stdio` | **JSONL IDE protocol** on stdin/stdout for editor extensions (hidden entry; used by VS Code). |
 
+**Files with a prompt (multimodal):** attach one or more files Gemini can read (images, **PDF**, audio, video, plain text, etc.) for that message only:
+
+```bash
+gemcode -C . --attach ./screenshot.png "Why does this UI look wrong?"
+gemcode -C . --attach ./report.pdf "What are the main conclusions?"
+gemcode -C . --image a.png --image b.png "Compare these two layouts"
+```
+
+(`--image` is an alias of `--attach`.)
+
+In the **REPL / TUI**, queue files for the **next** message: `/attach path/to.pdf` (repeat for multiple), then type your question. Use `/attach list`, `/attach clear`. Aliases: `/image`, `/img`, `/file`.
+
 ---
 
 ## Main CLI flags
@@ -110,6 +172,7 @@ Non-interactive environments (CI, pipes) must set `GOOGLE_API_KEY` explicitly an
 | `--tool-combination-mode` | Gemini 3 **tool context circulation**: `deep_research\|always\|never\|auto`. |
 | `--mcp` | Load MCP toolsets from `.gemcode/mcp.json` (requires `pip install -e ".[mcp]"`). |
 | `--max-llm-calls` | Cap model↔tool iterations (`RunConfig.max_llm_calls`). |
+| `--attach PATH` | Attach file(s) for **this** CLI message only (repeat flag). Gemini-supported MIME (e.g. images, PDF, audio, video). Default max ~20 MiB each (`GEMCODE_MAX_ATTACHMENT_BYTES`). Alias: `--image`. |
 
 Kaira and `live-audio` accept overlapping options (project root, `--yes`, research/embeddings, etc.); run `gemcode kaira -h` / `gemcode live-audio -h` for full lists.
 
@@ -122,6 +185,7 @@ State is **project-local** (unless noted).
 | Path / artifact | Purpose |
 |-----------------|---------|
 | `sessions.sqlite` | ADK session service: conversation history for `--session` ids. |
+| `GEMCODE_MEMORY.md`, `GEMCODE_USER.md` | **Curated memory** (see [Curated memory](#curated-memory)). |
 | `audit.log` | JSONL audit: tool usage, model usage, terminal reasons, optional tool-use summaries. |
 | `tool-results/` | Offloaded large tool outputs; references like `tool_result:<sha256>`. |
 | `artifacts/` | File artifacts (ADK `FileArtifactService`). |
@@ -232,10 +296,16 @@ In interactive mode, lines starting with `/` are **slash commands** (see `repl_c
 
 | Command | Purpose |
 |---------|---------|
+| `/attach <path>`, `/attach list`, `/attach clear` | Queue file(s) for the **next** message (Gemini multimodal). Aliases: **`/image`**, **`/img`**, **`/file`**. |
+| `/trust`, `/trust on`, `/trust off` | Show, grant, or revoke **workspace trust** for the project root (stored in `~/.gemcode/trust.json`; required for file/shell/git tools). |
 | `/init` \| `/init force` | Analyze the repo and generate or overwrite `GEMINI.md`. |
 | `/cost` | Token usage and estimated cost for the session. |
 | `/notes`, `/notes clear`, `/notes edit` | View, clear, or edit `.gemcode/notes.md`. |
-| `/create gemskill <name> [description]` | Scaffold `.gemcode/skills/<name>/SKILL.md`. |
+| `/create gemskill <name> [description]` | **Create** a new GemSkill (scaffold `SKILL.md`). |
+| `/gemskill <name>` | **Load** an existing skill into the **session system prompt** (until `/gemskill clear` or new session). |
+| `/gemskill list` \| `/gemskill clear` | List skills or unload all session-loaded skills. |
+| `/append gemskill <name> <request>` | **Iterate** the skill file: model edits `SKILL.md` per your request. |
+| `/curated` | Show **curated memory** snapshot (`.gemcode/GEMCODE_MEMORY.md`, `GEMCODE_USER.md`). |
 | `/style`, `/style <name>\|off` | List or activate **output styles** (`.gemcode/output-styles/*.md`). |
 | `/rules` | Show **rule** files from `.gemcode/rules/` (with path gating). |
 | `/diff`, `/diff last`, `/diff cp_…` | Git diff, or **checkpoint → workspace** diff. |
@@ -248,7 +318,7 @@ In interactive mode, lines starting with `/` are **slash commands** (see `repl_c
 | Command | Purpose |
 |---------|---------|
 | `/help` | Short help. |
-| `/status` | Model, capabilities, thinking, limits, risk/context telemetry. |
+| `/status` | Model, capabilities, thinking, limits, risk/context telemetry; **`loaded_skills`** (session-pinned GemSkills from `/gemskill`). |
 | `/config` | Dump active config fields. |
 | `/session`, `/session list`, `/session name`, `/session resume`, `/session new` | Session management; `/clear` aliases `/session new`. |
 | `/compact`, `/compact <focus>` | Force context compaction / summarization. |
@@ -256,6 +326,12 @@ In interactive mode, lines starting with `/` are **slash commands** (see `repl_c
 | `/context` | Context pressure and token breakdown (includes styles, rules, skills manifest, touched paths). |
 | `/audit [N]` | Tail of `audit.log`. |
 | `/tools` | Tool inventory for current config. |
+| `/tools smoke` | Declaration compile check only (lists failures). |
+| `/eval`, `/eval llm` | **Eval harness**: tool smoke + `pytest` if a `tests/` tree exists; optional LLM golden prompts (costs tokens). Writes `.gemcode/evals/last_eval.json`. |
+| `/autotune init <tag>` | Create git branch `autotune/<tag>` (requires a git repo). |
+| `/autotune eval`, `… llm` | Run eval and append `.gemcode/evals/autotune_ledger.jsonl`. |
+| `/login` | How to run **`gemcode login`** (API key is set outside the REPL). |
+| `/live-audio` | How to run **`gemcode live-audio`** (mic → Gemini Live). |
 | `/doctor` | Environment sanity check. |
 | `/version` | Version string. |
 | `/exit` | Leave the REPL. |
@@ -273,6 +349,7 @@ In interactive mode, lines starting with `/` are **slash commands** (see `repl_c
 |---------|---------|
 | `/computer`, `/computer on\|off`, `/computer url` | Browser automation (Playwright). |
 | `/research`, `/research on\|off` | Deep research tools. |
+| `/maps`, `/maps on\|off` | Maps **grounding** toggle (runner rebuild). |
 | `/embeddings on\|off` | Semantic search tool. |
 | `/caps`, `/caps …` | View or bulk-toggle capabilities. |
 | `/memory`, `/memory on\|off` | Persistent memory. |
@@ -307,7 +384,33 @@ The TUI (when `GEMCODE_TUI=1` and terminal supports it) provides **slash complet
 - **Discovery:** Only **metadata** (name + description) is preloaded into context for token efficiency. Full body loads **on demand** via `/skill <name>`, `/<name>`, or tools `load_skill` / `list_skills`.
 - **Built-in:** **`batch`** — parallel large-change workflow (map → units → `run_subtask` → verify). Exposed as `/batch <goal>`; not auto-invoked by the model (`disable_model_invocation`).
 
-Use **`/create gemskill <name>`** to scaffold a new skill directory.
+- **`/create gemskill <name>`** — create a new skill directory.
+- **`/gemskill <name>`** — pin the full skill body into the system prompt for this session.
+- **`/append gemskill <name> <text>`** — one-shot turn for the model to revise that skill on disk.
+
+---
+
+## Curated memory
+
+**Curated memory** is a small, **intentional** text layer separate from raw session logs or embedding retrieval:
+
+- **Project facts:** `.gemcode/GEMCODE_MEMORY.md` (conventions, commands, architecture notes the team wants the agent to see).
+- **User preferences (per project):** `.gemcode/GEMCODE_USER.md`.
+
+Legacy filenames **`.gemcode/MEMORY.md`** and **`.gemcode/USER.md`** are still read if the new names are absent.
+
+Content is **sanitized** before append (length and sensitivity heuristics). The REPL command **`/curated`** prints a bounded snapshot of what would be injected. When the **memory** capability is enabled, curated material can be combined with broader memory tools—see **`remember_fact`**, **`read_curated_memory`** in the [function tools](#function-tools-catalog) table.
+
+---
+
+## Workspace trust
+
+File, shell, and related tools require the **project root** to be **trusted**. Trust is stored in **`~/.gemcode/trust.json`** (path configurable via **`GEMCODE_HOME`**), not inside the repo.
+
+- **`/trust`** — show whether the current root is trusted.
+- **`/trust on`** / **`/trust off`** — grant or revoke trust for **`cfg.project_root`**.
+
+Without trust, mutating and shell tools remain blocked even if **`--yes`** is set—this reduces accidental execution when the cwd is wrong (e.g. home directory).
 
 ---
 

{gemcode-0.3.77 → gemcode-0.3.80}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "gemcode"
-version = "0.3.77"
+version = "0.3.80"
 description = "Local-first coding agent on Google Gemini + ADK"
 readme = "README.md"
 requires-python = ">=3.11"

{gemcode-0.3.77 → gemcode-0.3.80}/src/gemcode/agent.py

@@ -27,11 +27,49 @@ from gemcode.limits import make_before_model_limits_callback, make_before_model_
 from gemcode.thinking import build_thinking_config
 from gemcode.tools import build_function_tools
 from gemcode.tool_prompt_manifest import build_tool_manifest
-from gemcode.skills import build_skill_manifest_text
+from gemcode.skills import (
+  build_skill_manifest_text,
+  expand_skill_text,
+  list_supporting_files,
+  load_skill,
+)
 from gemcode.output_styles import build_output_style_section
 from gemcode.rules import build_rules_section
 
 
+def _build_session_loaded_skills_section(cfg: GemCodeConfig) -> str:
+  """Full bodies for GemSkills the user loaded with /gemskill (session-scoped)."""
+  names = list(getattr(cfg, "session_loaded_skill_names", None) or [])
+  if not names:
+    return ""
+  sid = getattr(cfg, "session_skill_expand_session_id", None) or ""
+  chunks: list[str] = []
+  seen: set[str] = set()
+  for raw in names:
+    sk_name = (raw or "").strip().lower()
+    if not sk_name or sk_name in seen:
+      continue
+    seen.add(sk_name)
+    s = load_skill(cfg.project_root, sk_name)
+    if s is None:
+      continue
+    expanded = expand_skill_text(s, arguments="", session_id=sid)
+    files = list_supporting_files(s)
+    head = f"### GemSkill: `/{s.meta.name}` (loaded for this session)\n\n"
+    chunk = head + expanded
+    if files:
+      chunk += f"\n\nSupporting files: {', '.join(files)}"
+    chunks.append(chunk)
+  if not chunks:
+    return ""
+  return (
+      "## Loaded GemSkills (this session)\n"
+      "The user explicitly loaded these skills with `/gemskill`. Follow their workflows "
+      "when the task matches their purpose; do not force them on unrelated requests.\n\n"
+      + "\n\n---\n\n".join(chunks)
+  )
+
+
 def build_global_instruction() -> str:
   """Global instruction applied to the entire agent tree (via ADK plugin)."""
   return (
@@ -888,6 +926,9 @@ You have two tools to persist project insights across sessions (auto-memory styl
   skill_manifest = build_skill_manifest_text(cfg.project_root)
   if skill_manifest:
     base = f"{base}\n\n{skill_manifest}"
+  loaded_skills = _build_session_loaded_skills_section(cfg)
+  if loaded_skills:
+    base = f"{base}\n\n{loaded_skills}"
   extra = _load_gemini_md(cfg.project_root)
   if extra.strip():
     return f"{base}\n\n## Project instructions (GEMINI.md)\n{extra}"

{gemcode-0.3.77 → gemcode-0.3.80}/src/gemcode/autotune.py

@@ -3,8 +3,9 @@ from __future__ import annotations
 import subprocess
 import time
 from pathlib import Path
-from typing import Any
+from typing import Any, Iterable
 
+from gemcode.config import GemCodeConfig
 from gemcode.evals.harness import run_eval_suite, write_eval_record
 
 
@@ -47,11 +48,24 @@ def init_autotune(*, project_root: Path, tag: str) -> dict[str, Any]:
   return {"status": "created", "branch": branch}
 
 
-def run_autotune_eval(*, project_root: Path, include_llm: bool, model: str | None = None) -> dict[str, Any]:
+def run_autotune_eval(
+  *,
+  project_root: Path,
+  include_llm: bool,
+  model: str | None = None,
+  session_cfg: GemCodeConfig | None = None,
+  extra_tools: Iterable[Any] | None = None,
+) -> dict[str, Any]:
   """
   Run eval suite and persist last result to .gemcode/evals/last_eval.json.
   """
-  res = run_eval_suite(project_root=project_root, include_llm=include_llm, model=model)
+  res = run_eval_suite(
+    project_root=project_root,
+    include_llm=include_llm,
+    model=model,
+    session_cfg=session_cfg,
+    extra_tools=extra_tools,
+  )
   meta = {
     "ts": time.time(),
     "git_sha": _git_head_sha(project_root),