autocontext 0.4.2__tar.gz → 0.4.4__tar.gz

{autocontext-0.4.2 → autocontext-0.4.4}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: autocontext
-Version: 0.4.2
+Version: 0.4.4
 Summary: autocontext control plane for iterative strategy evolution.
 Project-URL: Homepage, https://github.com/greyhaven-ai/autocontext
 Project-URL: Repository, https://github.com/greyhaven-ai/autocontext
@@ -23,6 +23,7 @@ Requires-Dist: fastapi>=0.116.1
 Requires-Dist: httpx>=0.28.1
 Requires-Dist: prime-sandboxes>=0.2.14
 Requires-Dist: pydantic>=2.11.0
+Requires-Dist: python-ulid>=3.0.0
 Requires-Dist: pyyaml>=6.0.2
 Requires-Dist: rich>=13.9.4
 Requires-Dist: typer>=0.16.0
@@ -60,7 +61,7 @@ The intended use is to hand the harness a real task in plain language, let it so
 pip install autocontext
 ```
 
-The current PyPI release line is `autocontext==0.4.2`.
+The current PyPI release line is `autocontext==0.4.4`.
 The PyPI package name is now `autocontext`. The CLI entrypoint remains `autoctx`.
 
 ## Working Directory
@@ -135,7 +136,7 @@ AUTOCONTEXT_CLAUDE_TIMEOUT=300 \
 uv run autoctx solve --description "improve customer-support replies for billing disputes" --gens 3
 ```
 
-For longer `autoctx judge` / `autoctx improve` prompts on `claude-cli`, use `--timeout <seconds>` or set `AUTOCONTEXT_CLAUDE_TIMEOUT`.
+For longer live prompts, `autoctx solve`, `autoctx judge`, and `autoctx improve` all accept `--timeout <seconds>`. `autoctx solve` also accepts `--generation-time-budget <seconds>` to cap per-generation solve runtime. You can still use provider env vars such as `AUTOCONTEXT_CLAUDE_TIMEOUT` or `AUTOCONTEXT_PI_TIMEOUT`.
 
 Run with Codex CLI (`codex exec` via a local authenticated Codex runtime):
 
@@ -157,14 +158,20 @@ uv run autoctx solve --description "improve customer-support replies for billing
 
 `autoctx investigate` now ships as a first-class Python CLI surface as well. It uses the architect runtime for investigation-spec synthesis and the analyst runtime for hypothesis generation, so role-routing overrides apply there too.
 
-Run with Pi RPC (remote Pi agent via HTTP):
+Run with Pi RPC (local Pi subprocess using `pi --mode rpc` JSONL):
 
 ```bash
 AUTOCONTEXT_AGENT_PROVIDER=pi-rpc \
-AUTOCONTEXT_PI_RPC_ENDPOINT=http://localhost:3284 \
+AUTOCONTEXT_PI_COMMAND=pi \
 uv run autoctx solve --description "improve customer-support replies for billing disputes" --gens 3
 ```
 
+For deterministic evals where Pi should ignore repo-local `AGENTS.md` / `CLAUDE.md`, add:
+
+```bash
+AUTOCONTEXT_PI_NO_CONTEXT_FILES=true
+```
+
 Run with Hermes (via OpenAI-compatible gateway):
 
 ```bash

{autocontext-0.4.2 → autocontext-0.4.4}/README.md

@@ -10,7 +10,7 @@ The intended use is to hand the harness a real task in plain language, let it so
 pip install autocontext
 ```
 
-The current PyPI release line is `autocontext==0.4.2`.
+The current PyPI release line is `autocontext==0.4.4`.
 The PyPI package name is now `autocontext`. The CLI entrypoint remains `autoctx`.
 
 ## Working Directory
@@ -85,7 +85,7 @@ AUTOCONTEXT_CLAUDE_TIMEOUT=300 \
 uv run autoctx solve --description "improve customer-support replies for billing disputes" --gens 3
 ```
 
-For longer `autoctx judge` / `autoctx improve` prompts on `claude-cli`, use `--timeout <seconds>` or set `AUTOCONTEXT_CLAUDE_TIMEOUT`.
+For longer live prompts, `autoctx solve`, `autoctx judge`, and `autoctx improve` all accept `--timeout <seconds>`. `autoctx solve` also accepts `--generation-time-budget <seconds>` to cap per-generation solve runtime. You can still use provider env vars such as `AUTOCONTEXT_CLAUDE_TIMEOUT` or `AUTOCONTEXT_PI_TIMEOUT`.
 
 Run with Codex CLI (`codex exec` via a local authenticated Codex runtime):
 
@@ -107,14 +107,20 @@ uv run autoctx solve --description "improve customer-support replies for billing
 
 `autoctx investigate` now ships as a first-class Python CLI surface as well. It uses the architect runtime for investigation-spec synthesis and the analyst runtime for hypothesis generation, so role-routing overrides apply there too.
 
-Run with Pi RPC (remote Pi agent via HTTP):
+Run with Pi RPC (local Pi subprocess using `pi --mode rpc` JSONL):
 
 ```bash
 AUTOCONTEXT_AGENT_PROVIDER=pi-rpc \
-AUTOCONTEXT_PI_RPC_ENDPOINT=http://localhost:3284 \
+AUTOCONTEXT_PI_COMMAND=pi \
 uv run autoctx solve --description "improve customer-support replies for billing disputes" --gens 3
 ```
 
+For deterministic evals where Pi should ignore repo-local `AGENTS.md` / `CLAUDE.md`, add:
+
+```bash
+AUTOCONTEXT_PI_NO_CONTEXT_FILES=true
+```
+
 Run with Hermes (via OpenAI-compatible gateway):
 
 ```bash

autocontext-0.4.4/assets/whats_new.txt

@@ -0,0 +1,5 @@
+Pi RPC and solve runtime budgets hardened for longer live-agent runs
+Custom scenario registry diagnostics and spec-to-scenario auto-materialization
+Structured agent-task JSON payloads now validate and render safely
+TypeScript new-scenario and improve fallbacks preserve family semantics
+Generated scenario solve/export paths keep family-specific signals intact

{autocontext-0.4.2 → autocontext-0.4.4}/docs/agent-integration.md

@@ -32,6 +32,7 @@ autoctx train --scenario grid_ctf --data data.jsonl --json
 ```
 
 **Contract:**
+
 - **stdout** receives the JSON payload (one JSON object per line).
 - **stderr** receives errors in the format `{"error": "description"}`.
 - **Exit code 0** means the command succeeded. The JSON payload is on stdout.
@@ -50,6 +51,7 @@ autoctx run --scenario my_agent_task --gens 3 --json
 ```
 
 JSON output shape:
+
 ```json
 {
   "run_id": "my_run",
@@ -67,6 +69,7 @@ autoctx status <run_id> --json
 ```
 
 JSON output shape:
+
 ```json
 {
   "run_id": "abc123",
@@ -92,6 +95,7 @@ autoctx list --json
 ```
 
 Returns an array of run summaries:
+
 ```json
 [
   {
@@ -157,6 +161,7 @@ autoctx export --scenario grid_ctf --output pkg.json --json
 ```
 
 JSON output shape:
+
 ```json
 {
   "scenario": "grid_ctf",
@@ -174,6 +179,7 @@ autoctx train --scenario grid_ctf --data training.jsonl --time-budget 300 --json
 ```
 
 JSON output shape:
+
 ```json
 {
   "scenario": "grid_ctf",
@@ -192,6 +198,7 @@ autoctx import-package --file grid_ctf_package.json --json
 ```
 
 JSON output shape:
+
 ```json
 {
   "scenario_name": "grid_ctf",
@@ -262,12 +269,14 @@ AUTOCONTEXT_PI_COMMAND=pi \
 AUTOCONTEXT_PI_TIMEOUT=120 \
 autoctx run --scenario my_task --json
 
-# Pi RPC (Pi agent via HTTP RPC — supports session persistence)
+# Pi RPC (Pi subprocess via `pi --mode rpc` JSONL)
 AUTOCONTEXT_AGENT_PROVIDER=pi-rpc \
-AUTOCONTEXT_PI_RPC_ENDPOINT=http://localhost:3284 \
-AUTOCONTEXT_PI_RPC_API_KEY=your-key \
+AUTOCONTEXT_PI_COMMAND=pi \
 autoctx run --scenario my_task --json
 
+# Optional: keep Pi deterministic by ignoring AGENTS.md / CLAUDE.md context files
+AUTOCONTEXT_PI_NO_CONTEXT_FILES=true
+
 # Role-scoped override: competitor uses a separate gateway/key
 AUTOCONTEXT_AGENT_PROVIDER=anthropic \
 ANTHROPIC_API_KEY=sk-ant-primary \
@@ -281,55 +290,62 @@ autoctx run --scenario my_task --json
 
 Key environment variables:
 
-| Variable | Purpose |
-|---|---|
-| `AUTOCONTEXT_AGENT_PROVIDER` | Agent provider: `anthropic`, `openai-compatible`, `ollama`, `vllm`, `pi`, `pi-rpc`, `deterministic` |
-| `AUTOCONTEXT_AGENT_API_KEY` | Global agent API key override (or use provider-native env vars such as `ANTHROPIC_API_KEY`) |
-| `AUTOCONTEXT_AGENT_BASE_URL` | Global base URL for OpenAI-compatible agent endpoints |
-| `AUTOCONTEXT_COMPETITOR_API_KEY` / `AUTOCONTEXT_COMPETITOR_BASE_URL` | Optional competitor-specific credential and endpoint override |
-| `AUTOCONTEXT_ANALYST_API_KEY` / `AUTOCONTEXT_ANALYST_BASE_URL` | Optional analyst-specific credential and endpoint override |
-| `AUTOCONTEXT_COACH_API_KEY` / `AUTOCONTEXT_COACH_BASE_URL` | Optional coach-specific credential and endpoint override |
-| `AUTOCONTEXT_ARCHITECT_API_KEY` / `AUTOCONTEXT_ARCHITECT_BASE_URL` | Optional architect-specific credential and endpoint override |
-| `AUTOCONTEXT_JUDGE_PROVIDER` | Judge provider (defaults to `anthropic`) |
-| `AUTOCONTEXT_JUDGE_API_KEY` | API key for the judge provider |
-| `AUTOCONTEXT_JUDGE_BASE_URL` | Base URL for OpenAI-compatible judge endpoints |
-| `AUTOCONTEXT_JUDGE_MODEL` | Override judge model name |
-| `AUTOCONTEXT_CLAUDE_MODEL` | Claude CLI model alias (default: `sonnet`) |
-| `AUTOCONTEXT_CLAUDE_TIMEOUT` | Claude CLI execution timeout in seconds (default: 120) |
-| `AUTOCONTEXT_MODEL_COMPETITOR` | Override competitor agent model |
-| `AUTOCONTEXT_DB_PATH` | SQLite database path |
-| `AUTOCONTEXT_PI_COMMAND` | Path to Pi CLI binary (default: `pi`) |
-| `AUTOCONTEXT_PI_TIMEOUT` | Pi CLI execution timeout in seconds (default: 120) |
-| `AUTOCONTEXT_PI_WORKSPACE` | Pi CLI working directory |
-| `AUTOCONTEXT_PI_MODEL` | Manual Pi model override (pins a specific checkpoint/path) |
-| `AUTOCONTEXT_PI_RPC_ENDPOINT` | Pi RPC server URL (default: `http://localhost:3284`) |
-| `AUTOCONTEXT_PI_RPC_API_KEY` | Pi RPC API key |
-| `AUTOCONTEXT_PI_RPC_SESSION_PERSISTENCE` | Persist Pi sessions across turns (default: `true`) |
+| Variable                                                             | Purpose                                                                                             |
+| -------------------------------------------------------------------- | --------------------------------------------------------------------------------------------------- |
+| `AUTOCONTEXT_AGENT_PROVIDER`                                         | Agent provider: `anthropic`, `openai-compatible`, `ollama`, `vllm`, `pi`, `pi-rpc`, `deterministic` |
+| `AUTOCONTEXT_AGENT_API_KEY`                                          | Global agent API key override (or use provider-native env vars such as `ANTHROPIC_API_KEY`)         |
+| `AUTOCONTEXT_AGENT_BASE_URL`                                         | Global base URL for OpenAI-compatible agent endpoints                                               |
+| `AUTOCONTEXT_COMPETITOR_API_KEY` / `AUTOCONTEXT_COMPETITOR_BASE_URL` | Optional competitor-specific credential and endpoint override                                       |
+| `AUTOCONTEXT_ANALYST_API_KEY` / `AUTOCONTEXT_ANALYST_BASE_URL`       | Optional analyst-specific credential and endpoint override                                          |
+| `AUTOCONTEXT_COACH_API_KEY` / `AUTOCONTEXT_COACH_BASE_URL`           | Optional coach-specific credential and endpoint override                                            |
+| `AUTOCONTEXT_ARCHITECT_API_KEY` / `AUTOCONTEXT_ARCHITECT_BASE_URL`   | Optional architect-specific credential and endpoint override                                        |
+| `AUTOCONTEXT_JUDGE_PROVIDER`                                         | Judge provider (defaults to `anthropic`)                                                            |
+| `AUTOCONTEXT_JUDGE_API_KEY`                                          | API key for the judge provider                                                                      |
+| `AUTOCONTEXT_JUDGE_BASE_URL`                                         | Base URL for OpenAI-compatible judge endpoints                                                      |
+| `AUTOCONTEXT_JUDGE_MODEL`                                            | Override judge model name                                                                           |
+| `AUTOCONTEXT_CLAUDE_MODEL`                                           | Claude CLI model alias (default: `sonnet`)                                                          |
+| `AUTOCONTEXT_CLAUDE_TIMEOUT`                                         | Claude CLI execution timeout in seconds (default: 120)                                              |
+| `AUTOCONTEXT_MODEL_COMPETITOR`                                       | Override competitor agent model                                                                     |
+| `AUTOCONTEXT_DB_PATH`                                                | SQLite database path                                                                                |
+| `AUTOCONTEXT_PI_COMMAND`                                             | Path to Pi CLI binary (default: `pi`)                                                               |
+| `AUTOCONTEXT_PI_TIMEOUT`                                             | Pi CLI execution timeout in seconds (default: 120)                                                  |
+| `AUTOCONTEXT_PI_WORKSPACE`                                           | Pi CLI working directory                                                                            |
+| `AUTOCONTEXT_PI_MODEL`                                               | Manual Pi model override (pins a specific checkpoint/path)                                          |
+| `AUTOCONTEXT_PI_NO_CONTEXT_FILES`                                    | Disable Pi context file loading (`AGENTS.md`, `CLAUDE.md`) for deterministic/eval-style runs        |
+| `AUTOCONTEXT_PI_RPC_ENDPOINT`                                        | Legacy compatibility field for older HTTP-based experiments; current Pi RPC runtime does not use it |
+| `AUTOCONTEXT_PI_RPC_API_KEY`                                         | Legacy compatibility field for older HTTP-based experiments; current Pi RPC runtime does not use it |
+| `AUTOCONTEXT_PI_RPC_SESSION_PERSISTENCE`                             | Toggle Pi session persistence when launching `pi --mode rpc` (default: `true`)                      |
 
 #### Pi CLI vs Pi RPC
 
 **Pi CLI** (`AUTOCONTEXT_AGENT_PROVIDER=pi`) invokes the `pi` binary in non-interactive `--print` mode for each agent turn. Best for:
+
 - Simple setups where Pi is installed locally
 - Stateless, one-shot agent executions
 - CI/testing environments
 
-**Pi RPC** (`AUTOCONTEXT_AGENT_PROVIDER=pi-rpc`) communicates with Pi via HTTP RPC. Best for:
-- Session persistence across multi-turn improvement loops
-- Branch-on-retry strategies (Pi creates branches for each retry)
-- Remote Pi instances running as a service
+**Pi RPC** (`AUTOCONTEXT_AGENT_PROVIDER=pi-rpc`) launches a local Pi subprocess in `--mode rpc` and exchanges LF-delimited JSONL over stdin/stdout. Best for:
+
+- Aligning autocontext with Pi's documented RPC protocol
+- Session-aware Pi runs when Pi session persistence is enabled
+- Local environments where the `pi` binary is available
 
 Both support **scenario-aware model handoff** when scenario context is available and no manual Pi model override is set. In that case, autocontext checks the distillation model registry for a scenario-specific checkpoint and routes to it automatically. If `AUTOCONTEXT_PI_MODEL` is set, that value is treated as a manual pin and used directly instead of consulting the registry. This enables the distill→deploy loop where a fine-tuned model is used for specific scenarios while still allowing operators to force a specific checkpoint when needed.
 
+Set `AUTOCONTEXT_PI_NO_CONTEXT_FILES=true` when you need Pi runs to ignore repository context files such as `AGENTS.md` and `CLAUDE.md`, which is especially useful for reproducible evaluations and other contamination-sensitive workflows.
+
 #### Hermes via OpenAI-Compatible Gateway
 
 Hermes exposes an OpenAI-compatible API server, so the fastest way to connect autocontext to Hermes is through the existing `openai-compatible` provider.
 
 **When to use the gateway path:**
+
 - You have a Hermes instance already running (local or remote)
 - You want the lowest-friction setup with standard chat-completions semantics
 - The OpenAI chat completions API surface is sufficient for your use case
 
 **Caveats:**
+
 - **Model naming**: Use the exact model name your Hermes server reports (e.g. `hermes-3-llama-3.1-8b`). Check `GET /v1/models` on your Hermes endpoint.
 - **Determinism**: Hermes temperature behavior may differ from OpenAI. Set `AUTOCONTEXT_JUDGE_TEMPERATURE=0.0` explicitly for reproducible evaluations.
 - **Memory/sessions**: The gateway path is stateless per-request. Hermes memory and tool configuration are server-side concerns, not managed by autocontext.
@@ -341,11 +357,13 @@ Hermes exposes an OpenAI-compatible API server, so the fastest way to connect au
 autocontext also supports Hermes directly through `AUTOCONTEXT_AGENT_PROVIDER=hermes`, which shells out to `hermes chat --query ...` instead of using the OpenAI-compatible gateway.
 
 **When to use the native runtime path:**
+
 - You want Hermes CLI behavior directly, including local SOUL/skill/tool configuration that Hermes applies in its own runtime
 - You want Hermes to run in a specific working directory via `AUTOCONTEXT_HERMES_WORKSPACE`
 - You want autocontext to call the local Hermes CLI without standing up a separate OpenAI-compatible server
 
 **Tradeoffs:**
+
 - **Still one-shot**: autocontext invokes Hermes in single-query mode. This is not the same thing as resuming a long-lived interactive Hermes chat session.
 - **CLI dependency**: The `hermes` binary must be installed and available on `PATH` (or configured via `AUTOCONTEXT_HERMES_COMMAND`).
 - **Endpoint overrides**: `AUTOCONTEXT_HERMES_BASE_URL` and `AUTOCONTEXT_HERMES_API_KEY` are forwarded into Hermes's provider env for custom OpenAI-compatible backends.
@@ -488,16 +506,16 @@ autoctx solve \
   --json | jq .
 ```
 
-`autoctx solve` is a synchronous CLI wrapper around the solve-on-demand pipeline. Use the server or MCP solve APIs if you need background job submission and later result retrieval from a long-lived process.
+`autoctx solve` is a synchronous CLI wrapper around the solve-on-demand pipeline. Use `--timeout <seconds>` when richer live prompts need a longer provider runtime window, and `--generation-time-budget <seconds>` when you want to cap per-generation solve runtime. Use the server or MCP solve APIs if you need background job submission and later result retrieval from a long-lived process.
 
 #### When to use which integration path
 
-| Path | Best for | Complexity |
-|------|----------|-----------|
-| **CLI-first** (this section) | Hermes agents driving `autoctx` via shell commands | Lowest |
-| **OpenAI-compatible provider** | autocontext calling Hermes for agent/judge completions | Low |
-| **MCP server** | Tool-catalog agents (Claude Code, MCP clients) | Medium |
-| **Native Hermes runtime** | autocontext calling the local Hermes CLI with Hermes-side workspace/skill context | Highest |
+| Path                           | Best for                                                                          | Complexity |
+| ------------------------------ | --------------------------------------------------------------------------------- | ---------- |
+| **CLI-first** (this section)   | Hermes agents driving `autoctx` via shell commands                                | Lowest     |
+| **OpenAI-compatible provider** | autocontext calling Hermes for agent/judge completions                            | Low        |
+| **MCP server**                 | Tool-catalog agents (Claude Code, MCP clients)                                    | Medium     |
+| **Native Hermes runtime**      | autocontext calling the local Hermes CLI with Hermes-side workspace/skill context | Highest    |
 
 The CLI-first path is recommended for getting started. Move to the gateway or native provider paths when you want autocontext to call Hermes instead of Hermes calling autocontext.
 
@@ -574,6 +592,7 @@ Once the server is running, invoke tools via the MCP protocol:
 ```
 
 Response:
+
 ```json
 {
   "content": [
@@ -619,6 +638,7 @@ This starts the autocontext MCP server on stdio when Hermes connects.
 For safe Hermes exposure, consider allowing tools by category:
 
 **Read-only (safe for any operator):**
+
 - `mcp_autocontext_list_scenarios` — Browse available scenarios
 - `mcp_autocontext_describe_scenario` — Get scenario details, rules, strategy interface
 - `mcp_autocontext_read_playbook` — Read accumulated strategy playbook
@@ -631,12 +651,14 @@ For safe Hermes exposure, consider allowing tools by category:
 - `mcp_autocontext_list_solved` — List scenarios with exported knowledge
 
 **Evaluation (stateless, safe):**
+
 - `mcp_autocontext_evaluate_output` — One-shot judge evaluation
 - `mcp_autocontext_validate_strategy` — Validate strategy JSON against scenario constraints
 - `mcp_autocontext_run_match` — Run a single match (deterministic)
 - `mcp_autocontext_run_tournament` — Run N matches with Elo scoring
 
 **Write operations (require operator trust):**
+
 - `mcp_autocontext_run_replay` — Replay a generation
 - `mcp_autocontext_export_skill` — Export strategy package
 - `mcp_autocontext_solve_scenario` — Launch a solve job (long-running, creates artifacts)
@@ -647,51 +669,65 @@ For safe Hermes exposure, consider allowing tools by category:
 Once configured, a Hermes agent can drive the full autocontext loop:
 
 **1. Discover scenarios:**
+
 ```
 Use autocontext_list_scenarios to see what's available.
 ```
+
 → Returns JSON array of scenario names with descriptions.
 
 **2. Inspect a scenario:**
+
 ```
 Use autocontext_describe_scenario with scenario_name="grid_ctf".
 ```
+
 → Returns rules, strategy interface, evaluation criteria, and scoring dimensions.
 
 **3. Validate a strategy:**
+
 ```
 Use autocontext_validate_strategy with scenario_name="grid_ctf" and
 strategy='{"aggression": 0.6, "defense": 0.4, "path_bias": 0.5}'.
 ```
+
 → Returns `{"valid": true, "reason": "ok"}` or validation errors.
 
 **4. Run a tournament:**
+
 ```
 Use autocontext_run_tournament with scenario_name="grid_ctf",
 strategy='{"aggression": 0.6, "defense": 0.4, "path_bias": 0.5}',
 matches=5.
 ```
+
 → Returns mean/best scores, Elo, wins/losses.
 
 **5. Read the playbook:**
+
 ```
 Use autocontext_read_playbook with scenario_name="grid_ctf".
 ```
+
 → Returns the accumulated playbook markdown (or sentinel if none exists).
 
 **6. Export knowledge:**
+
 ```
 Use autocontext_export_skill with scenario_name="grid_ctf".
 ```
+
 → Returns a portable skill package with playbook, lessons, best strategy.
 
 **7. Install the exported skill into Hermes:**
+
 ```
 Take the result from autocontext_export_skill, read result.skill_markdown and
 result.suggested_filename, and write the markdown into your Hermes skill directory.
 ```
 
 For raw MCP clients, `autocontext_export_skill` returns structured JSON that now includes:
+
 - `skill_markdown` — the rendered `SKILL.md` contents
 - `suggested_filename` — the recommended install filename, such as `grid-ctf-knowledge.md`
 
@@ -711,19 +747,20 @@ After writing the file, restart Hermes or reload its skills so the new knowledge
 All tools use the `autocontext_` prefix (e.g., `autocontext_list_scenarios`). This is deliberate — it prevents collisions in multi-MCP-server setups. In Hermes, the prefix is visible in tool discovery and helps distinguish autocontext tools from other MCP servers.
 
 **Known rough edges:**
+
 - Tool names are verbose — Hermes agents may need explicit instruction to use the `autocontext_` prefix
 - `autocontext_solve_scenario` is long-running and returns a `job_id`; poll with `autocontext_solve_status`
 - Sandbox tools require explicit create/destroy lifecycle management
 
 #### MCP vs CLI-First for Hermes
 
-| Aspect | MCP | CLI-first |
-|--------|-----|-----------|
-| **Setup** | Config in `mcp_servers` | Set env vars |
-| **Tool discovery** | Automatic (Hermes sees all tools) | Manual (`autoctx --help`) |
-| **Output format** | Structured MCP responses | `--json` stdout |
-| **Long-running jobs** | Poll via `autocontext_solve_status` | Poll via `autoctx status` |
-| **Best for** | Hermes agents with MCP support | Hermes agents with shell access |
+| Aspect                | MCP                                 | CLI-first                       |
+| --------------------- | ----------------------------------- | ------------------------------- |
+| **Setup**             | Config in `mcp_servers`             | Set env vars                    |
+| **Tool discovery**    | Automatic (Hermes sees all tools)   | Manual (`autoctx --help`)       |
+| **Output format**     | Structured MCP responses            | `--json` stdout                 |
+| **Long-running jobs** | Poll via `autocontext_solve_status` | Poll via `autoctx status`       |
+| **Best for**          | Hermes agents with MCP support      | Hermes agents with shell access |
 
 Use MCP when Hermes has native MCP client support and you want automatic tool discovery. Use CLI-first when you want simpler debugging or are scripting a workflow.
 

{autocontext-0.4.2 → autocontext-0.4.4}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "autocontext"
-version = "0.4.2"
+version = "0.4.4"
 description = "autocontext control plane for iterative strategy evolution."
 readme = "README.md"
 license = { text = "Apache-2.0" }
@@ -27,6 +27,7 @@ dependencies = [
   "httpx>=0.28.1",
   "prime-sandboxes>=0.2.14",
   "pydantic>=2.11.0",
+  "python-ulid>=3.0.0",
   "pyyaml>=6.0.2",
   "rich>=13.9.4",
   "typer>=0.16.0",

{autocontext-0.4.2 → autocontext-0.4.4}/src/autocontext/__init__.py

@@ -4,4 +4,4 @@ from autocontext.sdk import AutoContext
 
 __all__ = ["AutoContext", "__version__"]
 
-__version__ = "0.4.2"
+__version__ = "0.4.3"

{autocontext-0.4.2 → autocontext-0.4.4}/src/autocontext/agents/llm_client.py

@@ -207,19 +207,19 @@ class DeterministicDevClient(LanguageModelClient):
         del max_tokens, temperature, role
         self._rlm_turn_counter += 1
         if self._rlm_turn_counter == 1:
-            text = '<code>\nprint(type(answer))\nprint(answer)\n</code>'
+            text = "<code>\nprint(type(answer))\nprint(answer)\n</code>"
         elif self._rlm_turn_counter == 2:
             text = (
                 "<code>\n"
-                "answer[\"content\"] = (\n"
-                "    \"## Findings\\n\\n\"\n"
-                "    \"- Strategy balances offense/defense.\\n\\n\"\n"
-                "    \"## Root Causes\\n\\n\"\n"
-                "    \"- Moderate aggressiveness.\\n\\n\"\n"
-                "    \"## Actionable Recommendations\\n\\n\"\n"
-                "    \"- Increase defensive weight.\"\n"
+                'answer["content"] = (\n'
+                '    "## Findings\\n\\n"\n'
+                '    "- Strategy balances offense/defense.\\n\\n"\n'
+                '    "## Root Causes\\n\\n"\n'
+                '    "- Moderate aggressiveness.\\n\\n"\n'
+                '    "## Actionable Recommendations\\n\\n"\n'
+                '    "- Increase defensive weight."\n'
                 ")\n"
-                "answer[\"ready\"] = True\n"
+                'answer["ready"] = True\n'
                 "</code>"
             )
         else:
@@ -359,35 +359,44 @@ class DeterministicDevClient(LanguageModelClient):
             "name": "resource_balance",
             "display_name": "Resource Balance",
             "description": (
-                "A resource management scenario where agents balance mining, "
-                "defense, and trade to maximize colony growth."
+                "A resource management scenario where agents balance mining, defense, and trade to maximize colony growth."
             ),
             "strategy_interface_description": (
                 "Return JSON object with keys `mining`, `defense`, and `trade`, all floats in [0,1]. "
                 "Constraint: mining + defense + trade <= 2.0."
             ),
             "evaluation_criteria": (
-                "Optimize colony growth through efficient resource "
-                "allocation across mining, defense, and trade."
+                "Optimize colony growth through efficient resource allocation across mining, defense, and trade."
             ),
             "strategy_params": [
                 {
-                    "name": "mining", "description": "Investment in resource extraction",
-                    "min_value": 0.0, "max_value": 1.0, "default": 0.5,
+                    "name": "mining",
+                    "description": "Investment in resource extraction",
+                    "min_value": 0.0,
+                    "max_value": 1.0,
+                    "default": 0.5,
                 },
                 {
-                    "name": "defense", "description": "Investment in colony protection",
-                    "min_value": 0.0, "max_value": 1.0, "default": 0.4,
+                    "name": "defense",
+                    "description": "Investment in colony protection",
+                    "min_value": 0.0,
+                    "max_value": 1.0,
+                    "default": 0.4,
                 },
                 {
-                    "name": "trade", "description": "Investment in trade routes",
-                    "min_value": 0.0, "max_value": 1.0, "default": 0.5,
+                    "name": "trade",
+                    "description": "Investment in trade routes",
+                    "min_value": 0.0,
+                    "max_value": 1.0,
+                    "default": 0.5,
                 },
             ],
             "constraints": [
                 {
-                    "expression": "mining + defense + trade", "operator": "<=",
-                    "threshold": 2.0, "description": "total allocation must be <= 2.0",
+                    "expression": "mining + defense + trade",
+                    "operator": "<=",
+                    "threshold": 2.0,
+                    "description": "total allocation must be <= 2.0",
                 },
             ],
             "environment_variables": [
@@ -536,8 +545,7 @@ def build_client_from_settings(
         )
         if not api_key:
             raise ValueError(
-                "AUTOCONTEXT_ANTHROPIC_API_KEY or ANTHROPIC_API_KEY is required "
-                "when AUTOCONTEXT_AGENT_PROVIDER=anthropic"
+                "AUTOCONTEXT_ANTHROPIC_API_KEY or ANTHROPIC_API_KEY is required when AUTOCONTEXT_AGENT_PROVIDER=anthropic"
             )
         return AnthropicClient(api_key=api_key)
     if settings.agent_provider == "deterministic":
@@ -618,6 +626,7 @@ def build_client_from_settings(
             timeout=settings.pi_timeout,
             workspace=settings.pi_workspace,
             model=resolved_model,
+            no_context_files=settings.pi_no_context_files,
         )
         return RuntimeBridgeClient(PiCLIRuntime(pi_config))
     if settings.agent_provider == "pi-rpc":
@@ -626,8 +635,10 @@ def build_client_from_settings(
 
         rpc_config = PiRPCConfig(
             pi_command=settings.pi_command,
+            model=settings.pi_model,
             timeout=settings.pi_timeout,
             session_persistence=settings.pi_rpc_session_persistence,
+            no_context_files=settings.pi_no_context_files,
         )
         return RuntimeBridgeClient(PiRPCRuntime(rpc_config))
     if settings.agent_provider == "hermes":

{autocontext-0.4.2 → autocontext-0.4.4}/src/autocontext/agents/provider_bridge.py

@@ -328,6 +328,7 @@ def create_role_client(
             timeout=settings.pi_timeout,
             workspace=settings.pi_workspace,
             model=resolved_model,
+            no_context_files=settings.pi_no_context_files,
         )
         return RuntimeBridgeClient(PiCLIRuntime(pi_config))
 
@@ -336,8 +337,10 @@ def create_role_client(
 
         rpc_config = PiRPCConfig(
             pi_command=settings.pi_command,
+            model=settings.pi_model,
             timeout=settings.pi_timeout,
             session_persistence=settings.pi_rpc_session_persistence,
+            no_context_files=settings.pi_no_context_files,
         )
         return RuntimeBridgeClient(PiRPCRuntime(rpc_config))
 

{autocontext-0.4.2 → autocontext-0.4.4}/src/autocontext/banner.py

@@ -19,6 +19,33 @@ SYNC_BLOCK_START = "<!-- autocontext-readme-hero:start -->"
 SYNC_BLOCK_END = "<!-- autocontext-readme-hero:end -->"
 WHATS_NEW_BLOCK_START = "<!-- autocontext-whats-new:start -->"
 WHATS_NEW_BLOCK_END = "<!-- autocontext-whats-new:end -->"
+README_BADGES = (
+    (
+        "https://github.com/greyhaven-ai/autocontext/blob/main/LICENSE",
+        "https://img.shields.io/github/license/greyhaven-ai/autocontext",
+        "License",
+    ),
+    (
+        "https://github.com/greyhaven-ai/autocontext/stargazers",
+        "https://img.shields.io/github/stars/greyhaven-ai/autocontext",
+        "GitHub stars",
+    ),
+    (
+        "https://github.com/greyhaven-ai/autocontext/commits/main",
+        "https://img.shields.io/github/last-commit/greyhaven-ai/autocontext",
+        "Last commit",
+    ),
+    (
+        "https://pypi.org/project/autocontext/",
+        "https://img.shields.io/pypi/v/autocontext",
+        "PyPI version",
+    ),
+    (
+        "https://www.npmjs.com/package/autoctx",
+        "https://img.shields.io/npm/v/autoctx",
+        "npm version",
+    ),
+)
 
 
 def _assets_dir() -> Path:
@@ -138,12 +165,16 @@ def print_banner_rich() -> None:
 
 def render_readme_banner_block() -> str:
     """Render the synced README hero block."""
+    badges = "\n".join(f'  <a href="{href}"><img src="{image}" alt="{alt}"></a>' for href, image, alt in README_BADGES)
     return (
         f"{SYNC_BLOCK_START}\n"
         '<p align="center">\n'
         '  <img src="autocontext/assets/banner.svg" alt="autocontext ASCII banner" style="max-width: 100%; height: auto;" />\n'
         "</p>\n\n"
-        f'<p align="center"><strong>{TAGLINE}</strong></p>\n'
+        f'<p align="center"><strong>{TAGLINE}</strong></p>\n\n'
+        '<p align="center">\n'
+        f"{badges}\n"
+        "</p>\n\n"
         f"{SYNC_BLOCK_END}"
     )