agentless-mcp 0.1.0__py3-none-any.whl

agentless_mcp/__init__.py

@@ -0,0 +1 @@
+"""agentless-mcp: model-free tree-sitter repo-map and patch machinery."""

agentless_mcp/__main__.py

@@ -0,0 +1,6 @@
+"""Module entry point: ``python -m agentless_mcp`` runs the CLI bootstrap."""
+
+from agentless_mcp.bootstrap import cli_main
+
+if __name__ == "__main__":
+    raise SystemExit(cli_main())

agentless_mcp/adapters/__init__.py

@@ -0,0 +1 @@
+"""Adapters layer: CLI and MCP entry surfaces."""

agentless_mcp/adapters/cli/__init__.py

@@ -0,0 +1 @@
+"""CLI adapter: argparse tree; exit 0 ok, 1 domain, 2 usage-security."""

agentless_mcp/adapters/cli/formatting.py

@@ -0,0 +1,78 @@
+"""Output and exit-code conventions shared by every CLI subcommand.
+
+The split is deliberate and load-bearing for an agent driving this over Bash:
+**stdout carries the answer, stderr carries everything about the run**. A
+pipeline that captures stdout gets only the receipt, the banner and the view;
+a failure is on stderr with a non-zero exit, never interleaved into a view
+that would then parse as a shorter answer.
+
+Exit codes are the same three the plan fixes:
+
+* ``0`` -- the call answered, including answers that are legitimately empty.
+* ``1`` -- a domain failure: no such symbol, a file the grammar cannot parse,
+  a walk bound exceeded. The request made sense; the repository did not allow
+  it.
+* ``2`` -- usage or security: a bad flag, a missing repository root, a path
+  or root refused. The request itself was not admissible.
+"""
+
+import sys
+
+from agentless_mcp.application.repo_context import RepoContext
+from agentless_mcp.util.errors import (
+    AtlasError,
+    LanguageUnavailable,
+    RepoResolutionError,
+    SecurityRefusal,
+    WalkBoundExceeded,
+)
+
+EXIT_OK = 0
+EXIT_DOMAIN = 1
+EXIT_USAGE = 2
+
+
+def emit(text: str) -> None:
+    """Write an answer to stdout, exactly once, newline-terminated."""
+    sys.stdout.write(text if text.endswith("\n") else text + "\n")
+
+
+def note(text: str) -> None:
+    """Write a receipt or a summary to stderr, leaving stdout to the answer.
+
+    The write subcommands emit a unified diff or an equivalence key on stdout,
+    which a caller pipes straight into ``git apply`` or a comparison. Anything
+    describing the run -- which repository, which HEAD, how many edits landed
+    -- goes here, so that pipeline never has to be told to skip a header.
+    """
+    sys.stderr.write(text if text.endswith("\n") else text + "\n")
+
+
+def fail(message: str, code: int = EXIT_DOMAIN) -> int:
+    """Report a failure on stderr and return the exit code to propagate."""
+    sys.stderr.write(f"agentless-mcp: {message}\n")
+    return code
+
+
+def exit_code_for(error: AtlasError) -> int:
+    """Map a typed error onto its exit code.
+
+    Keyed on what the error *is* -- a refusal versus a degraded answer -- not
+    on which subcommand raised it, so a new subcommand inherits the mapping
+    instead of re-deciding it.
+    """
+    if isinstance(error, SecurityRefusal | RepoResolutionError):
+        return EXIT_USAGE
+    if isinstance(error, LanguageUnavailable | WalkBoundExceeded):
+        return EXIT_DOMAIN
+    return EXIT_DOMAIN
+
+
+def warn_about(ctx: RepoContext) -> None:
+    """Put a degraded repository state on stderr as well as in the receipt.
+
+    The receipt already carries the note, but a caller that pipes stdout into
+    a prompt would never see it there in time to act.
+    """
+    if ctx.note:
+        sys.stderr.write(f"agentless-mcp: {ctx.note}\n")