@gaia-minds/assistant-cli 0.1.1 → 0.2.0

package/README.md

@@ -36,7 +36,12 @@ See `assistant/README.md` for full runtime and release docs.
 
 - npm package is live: `@gaia-minds/assistant-cli@0.1.1`
 - Global CLI (`gaia`) supports onboarding, auth status, doctor, and dry-run loop execution
-- Runtime includes Gaia-native auth path with Codex CLI OAuth broker (`codex login --device-auth`)
+- `gaia onboard` now supports provider-guided setup:
+  - OpenRouter (API key + model selection)
+  - OpenAI (API key)
+  - Anthropic (API key)
+  - OpenAI Codex OAuth (`codex login --device-auth`)
+- Runtime reasoning provider supports `anthropic`, `openai`, and `openrouter`
 - Self-evolution loop runs in two tracks:
   - `assistant` (user-facing improvements)
   - `framework` (self-evolving engine improvements)

package/assistant/README.md

@@ -36,20 +36,43 @@ npm install -g .
 gaia doctor
 ```
 
-## OAuth Onboarding
+## Provider Onboarding
 
-Gaia uses its own local auth store for ChatGPT/Codex-style profiles.
-The default OAuth broker is Codex CLI (web/device flow):
+Run the guided onboarding wizard:
 
 ```bash
-npm run gaia -- auth login --source codex-cli --provider openai-codex
-npm run gaia -- auth status
+gaia onboard
+```
+
+The wizard lets you choose provider and connection style:
+
+1. `openrouter` -> API key + model selection
+2. `openai` -> API key
+3. `anthropic` -> API key
+4. `openai-codex` -> OAuth via Codex CLI
+
+Direct non-interactive examples:
+
+```bash
+# OpenRouter
+gaia onboard --provider openrouter --api-key "$OPENROUTER_API_KEY" --model openrouter/auto --yes
+
+# Anthropic
+gaia onboard --provider anthropic --api-key "$ANTHROPIC_API_KEY" --yes
+
+# OpenAI API key
+gaia onboard --provider openai --api-key "$OPENAI_API_KEY" --model gpt-4.1-mini --yes
+
+# OpenAI Codex OAuth
+gaia onboard --provider openai-codex --yes
 ```
 
-If you already authenticated with Codex CLI, import/link without re-running login:
+Gaia still supports explicit auth commands if you prefer manual control.
+For Codex OAuth:
 
 ```bash
-npm run gaia -- auth link --source codex-cli --provider openai-codex
+npm run gaia -- auth login --source codex-cli --provider openai-codex
+npm run gaia -- auth status
 ```
 
 Optional compatibility path (legacy): link from OpenClaw profile store.
@@ -59,6 +82,8 @@ Use `--source openclaw`.
 
 - OAuth tokens are stored in Gaia local state:
   `~/.gaia-assistant/auth-profiles.json` (or `$GAIA_ASSISTANT_HOME/auth-profiles.json`)
+- API keys can be stored in Gaia local secret store:
+  `~/.gaia-assistant/secrets.json` (or `$GAIA_ASSISTANT_HOME/secrets.json`)
 - Launcher config stores profile selection metadata in:
   `~/.gaia-assistant/config.json` (or `$GAIA_ASSISTANT_HOME/config.json`)
 - Never commit auth stores or local runtime state to git.
@@ -73,6 +98,39 @@ The evolution loop runs with two tracks:
 
 Default scheduling and policy live in `tools/agent-config.yml`.
 
+## Action Traces
+
+Gaia writes structured local traces to:
+
+- `~/.gaia-assistant/traces/actions.jsonl`
+
+Use:
+
+```bash
+gaia traces --last 5
+gaia traces --type file_read
+```
+
+Trace schema fields:
+
+- `id` (UUID)
+- `timestamp` (ISO-8601 UTC)
+- `action_type` (for example `chat_turn`, `note_capture`, `permission_decision`)
+- `input_summary` (redacted summary string)
+- `output_summary` (redacted summary string)
+- `duration_ms` (float)
+- `permission_level` (`safe`, `confirm`, `forbidden`)
+- `status` (`ok`, `error`, `blocked`)
+- `schema_version` (integer)
+- `metadata` (optional object for downstream tooling)
+
+Capability policy can be reviewed and overridden locally:
+
+```bash
+gaia capability list
+gaia capability set send_email confirm
+```
+
 ## Budget Policy
 
 Default budget split:
@@ -88,6 +146,36 @@ Expected environment variables for direct API mode:
 
 - `ANTHROPIC_API_KEY`
 - `OPENAI_API_KEY`
+- `OPENROUTER_API_KEY`
+
+Reasoning provider selection is configured by onboarding in launcher config,
+and can be overridden per run:
+
+```bash
+# one-off override from Gaia launcher
+gaia run --mode single --reasoning-provider openai --reasoning-model gpt-4.1-mini
+gaia run --mode single --reasoning-provider openrouter --reasoning-model openrouter/auto
+
+# npm/local clone equivalent
+npm run gaia -- run --mode single --reasoning-provider openai --reasoning-model gpt-4.1-mini
+npm run gaia -- run --mode single --reasoning-provider openrouter --reasoning-model openrouter/auto
+```
+
+OpenRouter quick setup:
+
+```bash
+export OPENROUTER_API_KEY="your-openrouter-key"
+gaia onboard --provider openrouter
+gaia run --mode single --reasoning-provider openrouter --reasoning-model openrouter/auto
+```
+
+OpenAI quick setup:
+
+```bash
+export OPENAI_API_KEY="your-openai-key"
+gaia onboard --provider openai --model gpt-4.1-mini
+gaia run --mode single --reasoning-provider openai --reasoning-model gpt-4.1-mini
+```
 
 Provider OAuth profile support is exposed through Gaia-native commands:
 
@@ -102,9 +190,33 @@ Direct Python fallback (if preferred):
 
 OpenClaw linking remains available as an optional compatibility source.
 
-Current limitation: the self-evolution loop planner currently uses Anthropic SDK
-for non-dry cycles. OAuth onboarding is in place so contributors can securely
-connect web auth profiles now while provider backends continue to evolve.
+The self-evolution loop planner supports Anthropic, OpenAI, and OpenRouter in
+non-dry runs. Tier-2 LLM alignment checks currently run only with Anthropic;
+when using OpenAI or OpenRouter, Tier-1 deterministic alignment checks still apply.
+
+## Agent Follow-Up
+
+For agents continuing this track, use contributor workflow from:
+
+- `skills/gaia-contributor/SKILL.md`
+- `skills/gaia-assistant-builder/SKILL.md`
+
+Recommended handoff protocol:
+
+1. Pull latest main and read this file + `tools/agent-config.yml`.
+2. Check open work first:
+   - `gh issue list --state open`
+   - `gh pr list --state open`
+3. Avoid duplication:
+   - `rg -n "<topic>" assistant tools infrastructure skills`
+4. Prefer small, reviewable PRs for runtime changes.
+5. Before pushing, run:
+   - `make check-all`
+   - `make test-smoke`
+   - `python3 -m py_compile tools/gaia-assistant.py tools/agent-loop.py tools/agent-alignment.py`
+6. Update `CHANGELOG.md` with meaningful behavior changes.
+7. If this is your first PR to Gaia, include Constitutional acknowledgment from
+   `skills/gaia-contributor/SKILL.md` in PR description.
 
 ## Maintainer Release Flow
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@gaia-minds/assistant-cli",
-  "version": "0.1.1",
+  "version": "0.2.0",
   "description": "npm-first CLI wrapper for the Gaia standalone assistant runtime",
   "author": "Gaia Minds Contributors",
   "license": "MIT",

package/tools/agent-config.yml

@@ -5,8 +5,8 @@
 # repository.  The agent runs on a schedule, picks tasks, validates them against
 # the Constitution, and opens PRs (or auto-executes low-risk actions).
 #
-# SECURITY NOTE: No secrets are stored here.  The Anthropic API key is read
-# from the ANTHROPIC_API_KEY environment variable at runtime.
+# SECURITY NOTE: No secrets are stored here. API keys are read from
+# environment variables at runtime.
 # =============================================================================
 
 # ---------------------------------------------------------------------------
@@ -18,13 +18,28 @@ agent:
   description: "Self-evolving autonomous agent for Gaia Minds"
 
 # ---------------------------------------------------------------------------
-# Claude API settings
+# Reasoning provider settings
 # ---------------------------------------------------------------------------
 reasoning:
-  model: "claude-sonnet-4-5-20250929"  # default model for reasoning
+  provider: "anthropic"  # anthropic | openai | openrouter
+  models:
+    anthropic: "claude-sonnet-4-5-20250929"
+    openai: "gpt-4.1-mini"
+    openrouter: "openrouter/auto"
   max_tokens: 4096
   temperature: 0.3  # low for consistent reasoning
-  # API key comes from ANTHROPIC_API_KEY env var, never stored in config
+  openai:
+    base_url: "https://api.openai.com/v1"
+    timeout_seconds: 120
+  openrouter:
+    base_url: "https://openrouter.ai/api/v1"
+    app_name: "gaia-minds-agent"
+    app_url: "https://github.com/Gaia-minds/gaia-minds"
+    timeout_seconds: 120
+  # API keys are read from env vars, never stored in config:
+  # - ANTHROPIC_API_KEY (for anthropic provider)
+  # - OPENAI_API_KEY (for openai provider)
+  # - OPENROUTER_API_KEY (for openrouter provider)
 
 # ---------------------------------------------------------------------------
 # Cycle settings