agent-cli 0.62.0__tar.gz → 0.63.0__tar.gz

{agent_cli-0.62.0 → agent_cli-0.63.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: agent-cli
-Version: 0.62.0
+Version: 0.63.0
 Summary: A suite of AI-powered command-line tools for text correction, audio transcription, and voice assistance.
 Project-URL: Homepage, https://github.com/basnijholt/agent-cli
 Author-email: Bas Nijholt <bas@nijho.lt>

{agent_cli-0.62.0 → agent_cli-0.63.0}/agent_cli/dev/cli.py

@@ -344,6 +344,19 @@ def _launch_editor(path: Path, editor: Editor) -> None:
         _warn(f"Could not open editor: {e}")
 
 
+def _write_prompt_to_worktree(worktree_path: Path, prompt: str) -> Path:
+    """Write the prompt to .claude/TASK.md in the worktree.
+
+    This makes the task description available to the spawned agent
+    and provides a record of what was requested.
+    """
+    claude_dir = worktree_path / ".claude"
+    claude_dir.mkdir(parents=True, exist_ok=True)
+    task_file = claude_dir / "TASK.md"
+    task_file.write_text(prompt + "\n")
+    return task_file
+
+
 def _format_env_prefix(env: dict[str, str]) -> str:
     """Format environment variables as shell prefix.
 
@@ -627,6 +640,11 @@ def new(  # noqa: PLR0912, PLR0915
             # Only warn if user explicitly requested direnv
             _warn("direnv not installed, skipping .envrc setup")
 
+    # Write prompt to worktree (makes task available to the spawned agent)
+    if prompt:
+        task_file = _write_prompt_to_worktree(result.path, prompt)
+        _success(f"Wrote task to {task_file.relative_to(result.path)}")
+
     # Resolve editor and agent
     resolved_editor = _resolve_editor(editor, editor_name, default_editor)
     resolved_agent = _resolve_agent(agent, agent_name, default_agent)
@@ -1039,6 +1057,11 @@ def start_agent(
     if not agent.is_available():
         _error(f"{agent.name} is not installed. Install from: {agent.install_url}")
 
+    # Write prompt to worktree (makes task available to the agent)
+    if prompt:
+        task_file = _write_prompt_to_worktree(wt.path, prompt)
+        _success(f"Wrote task to {task_file.relative_to(wt.path)}")
+
     merged_args = _merge_agent_args(agent, agent_args)
     agent_env = _get_agent_env(agent)
     _info(f"Starting {agent.name} in {wt.path}...")

{agent_cli-0.62.0 → agent_cli-0.63.0}/agent_cli/dev/skill/SKILL.md

@@ -47,8 +47,9 @@ agent-cli dev new <branch-name> --agent --prompt-file path/to/prompt.md
 This creates:
 1. A new git worktree with its own branch
 2. Runs project setup (installs dependencies)
-3. Opens a new terminal tab with an AI coding agent
-4. Passes your prompt to the agent
+3. Saves your prompt to `.claude/TASK.md` in the worktree (for reference)
+4. Opens a new terminal tab with an AI coding agent
+5. Passes your prompt to the agent
 
 **Important**: Use `--prompt-file` for prompts longer than a single line. The `--prompt` option passes text through the shell, which can cause issues with special characters (exclamation marks, dollar signs, backticks, quotes) in ZSH and other shells. Using `--prompt-file` avoids all shell quoting issues.
 

{agent_cli-0.62.0 → agent_cli-0.63.0}/agent_cli/memory/_retrieval.py

@@ -25,6 +25,7 @@ LOGGER = logging.getLogger(__name__)
 
 _DEFAULT_MMR_LAMBDA = 0.7
 _SUMMARY_ROLE = "summary"
+_MIN_MAX_EPSILON = 1e-8  # Avoid division by zero in min-max normalization
 
 
 def gather_relevant_existing_memories(
@@ -135,7 +136,7 @@ def retrieve_memory(
     include_summary: bool = True,
     mmr_lambda: float = _DEFAULT_MMR_LAMBDA,
     recency_weight: float = 0.2,
-    score_threshold: float = 0.35,
+    score_threshold: float | None = None,
     filters: dict[str, Any] | None = None,
 ) -> tuple[MemoryRetrieval, list[str]]:
     """Execute search + rerank + recency + MMR."""
@@ -161,8 +162,15 @@ def retrieve_memory(
             seen_ids.add(rec_id)
             raw_candidates.append(rec)
 
-    def _sigmoid(x: float) -> float:
-        return 1.0 / (1.0 + math.exp(-x))
+    def _min_max_normalize(scores: list[float]) -> list[float]:
+        """Normalize scores to 0-1 range using min-max scaling."""
+        if not scores:
+            return scores
+        min_score = min(scores)
+        max_score = max(scores)
+        if max_score - min_score < _MIN_MAX_EPSILON:
+            return [0.5] * len(scores)  # All scores equal
+        return [(s - min_score) / (max_score - min_score) for s in scores]
 
     def recency_score(meta: MemoryMetadata) -> float:
         dt = datetime.fromisoformat(meta.created_at)
@@ -176,10 +184,12 @@ def retrieve_memory(
     if raw_candidates:
         pairs = [(query, mem.content) for mem in raw_candidates]
         rr_scores = predict_relevance(reranker_model, pairs)
-        for mem, rr in zip(raw_candidates, rr_scores, strict=False):
-            relevance = _sigmoid(rr)
-            # Filter out low-relevance memories to reduce noise
-            if relevance < score_threshold:
+        # Normalize raw reranker scores to 0-1 range
+        normalized_scores = _min_max_normalize(rr_scores)
+
+        for mem, relevance in zip(raw_candidates, normalized_scores, strict=False):
+            # Filter out low-relevance memories if threshold is set
+            if score_threshold is not None and relevance < score_threshold:
                 continue
 
             recency = recency_score(mem.metadata)
@@ -235,7 +245,7 @@ async def augment_chat_request(
     include_global: bool = True,
     mmr_lambda: float = _DEFAULT_MMR_LAMBDA,
     recency_weight: float = 0.2,
-    score_threshold: float = 0.35,
+    score_threshold: float | None = None,
     filters: dict[str, Any] | None = None,
 ) -> tuple[ChatRequest, MemoryRetrieval | None, str, list[str]]:
     """Retrieve memory context and augment the chat request."""

{agent_cli-0.62.0 → agent_cli-0.63.0}/agent_cli/memory/api.py

@@ -30,7 +30,7 @@ def create_app(
     max_entries: int = 500,
     mmr_lambda: float = 0.7,
     recency_weight: float = 0.2,
-    score_threshold: float = 0.35,
+    score_threshold: float | None = None,
     enable_git_versioning: bool = True,
 ) -> FastAPI:
     """Create the FastAPI app for memory-backed chat."""

{agent_cli-0.62.0 → agent_cli-0.63.0}/agent_cli/memory/client.py

@@ -49,7 +49,7 @@ class MemoryClient:
         max_entries: int = 500,
         mmr_lambda: float = 0.7,
         recency_weight: float = 0.2,
-        score_threshold: float = 0.35,
+        score_threshold: float | None = None,
         start_watcher: bool = False,
         enable_git_versioning: bool = True,
     ) -> None:

{agent_cli-0.62.0 → agent_cli-0.63.0}/agent_cli/memory/engine.py

@@ -236,7 +236,7 @@ async def process_chat_request(
     max_entries: int = _DEFAULT_MAX_ENTRIES,
     mmr_lambda: float = _DEFAULT_MMR_LAMBDA,
     recency_weight: float = 0.2,
-    score_threshold: float = 0.35,
+    score_threshold: float | None = None,
     postprocess_in_background: bool = True,
     enable_git_versioning: bool = False,
     filters: dict[str, Any] | None = None,

{agent_cli-0.62.0 → agent_cli-0.63.0}/docs/architecture/memory.md

@@ -207,8 +207,8 @@ Executed synchronously during `augment_chat_request`.
 *   **Model:** `Xenova/ms-marco-MiniLM-L-6-v2` (ONNX, quantized).
 *   **Input:** Pairs of `(user_query, memory_content)`.
 *   **Output:** Raw logit score.
-*   **Normalization:** `relevance = sigmoid(raw_score)`.
-*   **Thresholding:** Candidates with `relevance < score_threshold` (default 0.35) are hard-pruned.
+*   **Normalization:** Min-max scaling across the candidate batch (best match = 1.0, worst = 0.0).
+*   **Thresholding:** If `score_threshold` is set, candidates below it are pruned. Default is `None` (no filtering).
 
 ### Step 3: Recency Scoring
 Combines semantic relevance with temporal proximity.
@@ -309,7 +309,7 @@ To replicate the system behavior, the following prompt strategies are required.
 | `max_entries` | `500` | Hard cap on memories per conversation. |
 | `mmr_lambda` | `0.7` | Diversity weighting (1.0 = pure relevance). |
 | `recency_weight` | `0.2` | Score weight for temporal proximity. |
-| `score_threshold` | `0.35` | Minimum semantic relevance to consider. |
+| `score_threshold` | `None` | Minimum semantic relevance to consider (no filtering by default). |
 | `enable_summarization` | `True` | Toggle for summary generation loop. |
 | `openai_base_url` | *required* | Base URL for LLM calls (can point to OpenAI-compatible proxies). |
 | `enable_git_versioning` | `True` | Toggle to enable/disable Git versioning of the memory store. |

{agent_cli-0.62.0 → agent_cli-0.63.0}/docs/commands/dev.md

@@ -61,8 +61,8 @@ agent-cli dev new [BRANCH] [OPTIONS]
 | `--with-editor` | Specific editor (cursor, vscode, zed, etc.) |
 | `--with-agent` | Specific agent (claude, codex, gemini, aider) |
 | `--agent-args` | Extra arguments to pass to the agent |
-| `--prompt`, `-p` | Initial prompt to pass to the AI agent |
-| `--prompt-file`, `-P` | Read initial prompt from a file (avoids shell quoting issues) |
+| `--prompt`, `-p` | Initial prompt to pass to the AI agent (saved to `.claude/TASK.md` in worktree) |
+| `--prompt-file`, `-P` | Read initial prompt from a file (saved to `.claude/TASK.md` in worktree) |
 | `--direnv` | Generate .envrc file for direnv (auto-detects venv) |
 | `--setup/--no-setup` | Run automatic project setup (default: enabled) |
 | `--copy-env/--no-copy-env` | Copy .env files from main repo (default: enabled) |
@@ -223,8 +223,8 @@ agent-cli dev agent NAME [--agent/-a AGENT] [--agent-args ARGS] [--prompt/-p PRO
 |--------|-------------|
 | `--agent`, `-a` | Specific agent (claude, codex, gemini, aider) |
 | `--agent-args` | Extra arguments to pass to the agent |
-| `--prompt`, `-p` | Initial prompt to pass to the AI agent |
-| `--prompt-file`, `-P` | Read initial prompt from a file (avoids shell quoting issues) |
+| `--prompt`, `-p` | Initial prompt to pass to the AI agent (saved to `.claude/TASK.md` in worktree) |
+| `--prompt-file`, `-P` | Read initial prompt from a file (saved to `.claude/TASK.md` in worktree) |
 
 **Examples:**