feat(lib): bundle MCP servers from .common/lib into factory/lib/
Copies source (no node_modules, dist, .tasks, .wiki, __pycache__) for: - projects-meta-mcp v2.25.0 (TypeScript/Node) - wiki-graph v0.3.1 (TypeScript/Node) - interns-mcp v0.3.3 (Python/FastMCP) .gitignore: exclude lib build artefacts (node_modules, dist, .venv, __pycache__, *.pyc) bootstrap.ps1: add MCP build step — npm install+build for TS servers, venv+pip for Python Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
This commit is contained in:
0
lib/interns-mcp/interns_mcp/interns/__init__.py
Normal file
0
lib/interns-mcp/interns_mcp/interns/__init__.py
Normal file
52
lib/interns-mcp/interns_mcp/interns/base.py
Normal file
52
lib/interns-mcp/interns_mcp/interns/base.py
Normal file
@@ -0,0 +1,52 @@
|
||||
"""Base intern protocol and response types."""
|
||||
|
||||
from __future__ import annotations
|
||||
|
||||
from dataclasses import dataclass, field
|
||||
from typing import Any
|
||||
|
||||
|
||||
@dataclass
|
||||
class InternResponse:
|
||||
text: str
|
||||
usage: dict[str, Any] = field(default_factory=dict)
|
||||
|
||||
|
||||
@dataclass
|
||||
class BlockedByPolicy:
|
||||
path: str
|
||||
pattern: str
|
||||
reason: str
|
||||
|
||||
|
||||
@dataclass
|
||||
class BudgetExceededError:
|
||||
tokens: int
|
||||
limit: int
|
||||
top_files: list[str] = field(default_factory=list)
|
||||
hint: str = "use compress=True or narrow paths"
|
||||
|
||||
|
||||
class Intern:
|
||||
id: str = ""
|
||||
description: str = ""
|
||||
endpoint: str | None = None
|
||||
model: str | None = None
|
||||
max_tokens: int = 4096
|
||||
temperature: float = 0.2
|
||||
system_prompt: str = ""
|
||||
context_budget: int = 120000
|
||||
|
||||
def __init__(self, config: dict[str, Any] | None = None, client: Any = None, safety: Any = None):
|
||||
self.client = client
|
||||
self.safety = safety
|
||||
if config:
|
||||
self._apply_config(config)
|
||||
|
||||
def _apply_config(self, config: dict[str, Any]) -> None:
|
||||
for k, v in config.items():
|
||||
if hasattr(self, k):
|
||||
setattr(self, k, v)
|
||||
|
||||
def run(self, paths: list[str], question: str, **kwargs: Any) -> InternResponse | BlockedByPolicy | BudgetExceededError:
|
||||
raise NotImplementedError
|
||||
46
lib/interns-mcp/interns_mcp/interns/bulk_text_read.py
Normal file
46
lib/interns-mcp/interns_mcp/interns/bulk_text_read.py
Normal file
@@ -0,0 +1,46 @@
|
||||
"""Bulk text read intern — read N files and answer a focused question."""
|
||||
|
||||
from __future__ import annotations
|
||||
|
||||
from pathlib import Path
|
||||
from typing import Any
|
||||
|
||||
from .base import BlockedByPolicy, Intern, InternResponse
|
||||
|
||||
|
||||
class BulkTextRead(Intern):
|
||||
id = "bulk_text_read"
|
||||
description = "Read N files and answer a focused question. Returns concise summary."
|
||||
|
||||
def run(self, paths: list[str], question: str, **kwargs: Any) -> InternResponse | BlockedByPolicy:
|
||||
contents: list[str] = []
|
||||
for p in paths:
|
||||
text = Path(p).read_text(encoding="utf-8", errors="replace")
|
||||
contents.append(f"--- {p} ---\n{text}")
|
||||
|
||||
combined = "\n\n".join(contents)
|
||||
prompt = f"Files:\n{combined}\n\nQuestion: {question}"
|
||||
|
||||
if self.client is None:
|
||||
return InternResponse(text="No client configured", usage={})
|
||||
|
||||
extra_body = kwargs.get("extra_body", {})
|
||||
resp = self.client.chat.completions.create(
|
||||
model=self.model or "deepseek-v4-flash",
|
||||
messages=[
|
||||
{"role": "system", "content": self.system_prompt},
|
||||
{"role": "user", "content": prompt},
|
||||
],
|
||||
max_tokens=kwargs.get("max_tokens", self.max_tokens),
|
||||
temperature=self.temperature,
|
||||
extra_body=extra_body if extra_body else None,
|
||||
)
|
||||
|
||||
choice = resp.choices[0]
|
||||
return InternResponse(
|
||||
text=choice.message.content or "",
|
||||
usage={
|
||||
"tokens_in": resp.usage.prompt_tokens if resp.usage else 0,
|
||||
"tokens_out": resp.usage.completion_tokens if resp.usage else 0,
|
||||
},
|
||||
)
|
||||
104
lib/interns-mcp/interns_mcp/interns/grep_audit.py
Normal file
104
lib/interns-mcp/interns_mcp/interns/grep_audit.py
Normal file
@@ -0,0 +1,104 @@
|
||||
"""Grep audit intern — deterministic matrix of N paths x M patterns. No LLM call."""
|
||||
|
||||
from __future__ import annotations
|
||||
|
||||
import json
|
||||
import re
|
||||
from pathlib import Path
|
||||
from typing import Any, Callable, Literal
|
||||
|
||||
from .base import Intern, InternResponse
|
||||
|
||||
|
||||
class GrepAudit(Intern):
|
||||
id = "grep_audit"
|
||||
description = (
|
||||
"Deterministic grep matrix over N paths x M patterns (substring or regex). "
|
||||
"No LLM call, no endpoint cost. Returns markdown table (default) or JSON."
|
||||
)
|
||||
|
||||
def run(
|
||||
self,
|
||||
paths: list[str],
|
||||
patterns: list[str | dict[str, Any]],
|
||||
output: Literal["table", "json"] = "table",
|
||||
case_sensitive: bool = True,
|
||||
**_kwargs: Any,
|
||||
) -> InternResponse:
|
||||
compiled = _compile_patterns(patterns, case_sensitive)
|
||||
|
||||
rows: list[dict[str, Any]] = []
|
||||
matches_total = 0
|
||||
for p in paths:
|
||||
try:
|
||||
text = Path(p).read_text(encoding="utf-8", errors="replace")
|
||||
except (FileNotFoundError, PermissionError, IsADirectoryError, OSError) as exc:
|
||||
rows.append(
|
||||
{
|
||||
"path": p,
|
||||
"matches": {c["name"]: None for c in compiled},
|
||||
"error": str(exc),
|
||||
}
|
||||
)
|
||||
continue
|
||||
row_matches: dict[str, bool] = {}
|
||||
for c in compiled:
|
||||
hit = bool(c["matcher"](text))
|
||||
row_matches[c["name"]] = hit
|
||||
if hit:
|
||||
matches_total += 1
|
||||
rows.append({"path": p, "matches": row_matches})
|
||||
|
||||
if output == "json":
|
||||
rendered = json.dumps({"rows": rows}, ensure_ascii=False, indent=2)
|
||||
else:
|
||||
rendered = _render_table(rows, [c["name"] for c in compiled])
|
||||
|
||||
return InternResponse(
|
||||
text=rendered,
|
||||
usage={
|
||||
"files_scanned": len(paths),
|
||||
"patterns_evaluated": len(compiled),
|
||||
"matches_total": matches_total,
|
||||
},
|
||||
)
|
||||
|
||||
|
||||
def _compile_patterns(
|
||||
patterns: list[str | dict[str, Any]], case_sensitive: bool
|
||||
) -> list[dict[str, Any]]:
|
||||
out: list[dict[str, Any]] = []
|
||||
for p in patterns:
|
||||
if isinstance(p, str):
|
||||
out.append({"name": p, "matcher": _substring_matcher(p, case_sensitive)})
|
||||
continue
|
||||
name = p.get("name", p["pattern"])
|
||||
if p.get("regex"):
|
||||
flags = 0 if case_sensitive else re.IGNORECASE
|
||||
compiled = re.compile(p["pattern"], flags)
|
||||
out.append({"name": name, "matcher": compiled.search})
|
||||
else:
|
||||
out.append(
|
||||
{"name": name, "matcher": _substring_matcher(p["pattern"], case_sensitive)}
|
||||
)
|
||||
return out
|
||||
|
||||
|
||||
def _substring_matcher(needle: str, case_sensitive: bool) -> Callable[[str], bool]:
|
||||
if case_sensitive:
|
||||
return lambda text: needle in text
|
||||
lowered = needle.lower()
|
||||
return lambda text: lowered in text.lower()
|
||||
|
||||
|
||||
def _render_table(rows: list[dict[str, Any]], names: list[str]) -> str:
|
||||
header = "| Path | " + " | ".join(names) + " |"
|
||||
sep = "|" + "---|" * (len(names) + 1)
|
||||
lines = [header, sep]
|
||||
for row in rows:
|
||||
cells: list[str] = []
|
||||
for n in names:
|
||||
v = row["matches"].get(n)
|
||||
cells.append("⚠️" if v is None else ("✅" if v else "❌"))
|
||||
lines.append(f"| `{row['path']}` | " + " | ".join(cells) + " |")
|
||||
return "\n".join(lines)
|
||||
164
lib/interns-mcp/interns_mcp/interns/repo_read.py
Normal file
164
lib/interns-mcp/interns_mcp/interns/repo_read.py
Normal file
@@ -0,0 +1,164 @@
|
||||
"""Repo read intern — pack repo with repomix, answer question with LLM."""
|
||||
|
||||
from __future__ import annotations
|
||||
|
||||
import os
|
||||
import shutil
|
||||
import subprocess
|
||||
import tempfile
|
||||
from pathlib import Path
|
||||
from typing import Any
|
||||
|
||||
from .base import BudgetExceededError, Intern, InternResponse
|
||||
from ..safety import always_ask_globs
|
||||
|
||||
|
||||
# Conservative token estimate: chars/4 overestimates for repomix XML (tag overhead),
|
||||
# so BudgetExceededError fires early rather than late. Real enforcement is on the
|
||||
# endpoint side — this is a client-side safety net, not a precision counter.
|
||||
CHARS_PER_TOKEN = 4
|
||||
|
||||
|
||||
def _estimate_tokens(text: str) -> int:
|
||||
return len(text) // CHARS_PER_TOKEN
|
||||
|
||||
|
||||
def _top_files_by_size(packed: str, n: int = 5) -> list[str]:
|
||||
"""Return the n largest files in repomix XML output, by block content size.
|
||||
|
||||
Parses each ``<file path="X">...</file>`` block and measures the byte length
|
||||
of its content, so BudgetExceededError.top_files points the user at the real
|
||||
overflow culprits — not the first files in document order.
|
||||
"""
|
||||
sizes: list[tuple[int, str]] = []
|
||||
pos = 0
|
||||
while True:
|
||||
open_idx = packed.find('<file ', pos)
|
||||
if open_idx == -1:
|
||||
break
|
||||
close_idx = packed.find('</file>', open_idx)
|
||||
if close_idx == -1:
|
||||
break
|
||||
path_start = packed.find('path="', open_idx) + 6
|
||||
path_end = packed.find('"', path_start)
|
||||
path = packed[path_start:path_end]
|
||||
# Content sits between the opening tag's '>' and the closing '</file>'.
|
||||
content_start = packed.find('>', open_idx) + 1
|
||||
content_size = close_idx - content_start
|
||||
sizes.append((content_size, path))
|
||||
pos = close_idx + len('</file>')
|
||||
|
||||
sizes.sort(key=lambda x: x[0], reverse=True)
|
||||
return [path for _, path in sizes[:n]]
|
||||
|
||||
|
||||
class RepoRead(Intern):
|
||||
id = "repo_read"
|
||||
description = "Pack directory/repo with repomix and answer a question. Cheap repo comprehension."
|
||||
|
||||
def run(
|
||||
self,
|
||||
paths: list[str],
|
||||
question: str,
|
||||
compress: bool = False,
|
||||
**kwargs: Any,
|
||||
) -> InternResponse | BudgetExceededError:
|
||||
if not paths:
|
||||
return InternResponse(text="No paths provided", usage={})
|
||||
|
||||
# Build repomix args: pass all paths, not just first.
|
||||
ignore_globs = always_ask_globs()
|
||||
|
||||
# Resolve npx via PATH. On Windows, subprocess.run(shell=False) does not
|
||||
# expand PATHEXT, so passing bare "npx" fails to find npx.cmd; shutil.which
|
||||
# finds the .cmd (or .exe / shell wrapper on POSIX) and returns a full path
|
||||
# that CreateProcess / execve can dispatch directly.
|
||||
npx = shutil.which("npx") or shutil.which("npx.cmd")
|
||||
if npx is None:
|
||||
return InternResponse(text="npx not found - install Node.js", usage={})
|
||||
|
||||
# Use mkstemp to avoid Windows file-sharing issues with NamedTemporaryFile.
|
||||
# os.close(fd) releases the handle before repomix writes to the path.
|
||||
fd, tmp_path = tempfile.mkstemp(suffix=".xml")
|
||||
os.close(fd)
|
||||
err_fd, err_path = tempfile.mkstemp(suffix=".stderr")
|
||||
os.close(err_fd)
|
||||
try:
|
||||
args = [
|
||||
npx, "-y", "repomix@latest",
|
||||
*paths,
|
||||
"--output", tmp_path,
|
||||
"--style", "xml",
|
||||
]
|
||||
if compress:
|
||||
args.append("--compress")
|
||||
for glob in ignore_globs:
|
||||
args += ["--ignore", glob]
|
||||
|
||||
try:
|
||||
# Redirect repomix's stdio to FILES/DEVNULL, never PIPES. capture_output
|
||||
# spawns reader threads that communicate() joins; if any process in the
|
||||
# child tree (an orphaned `node` grandchild, or — under concurrent launches
|
||||
# on Windows — a sibling subprocess that inherited the pipe handle) keeps a
|
||||
# pipe write-end open, that join blocks FOREVER, past the timeout, wedging
|
||||
# the worker thread. repomix's real output already goes to --output, so its
|
||||
# stdout is unused; stderr is captured to a file for error reporting only.
|
||||
with open(err_path, "w", encoding="utf-8", errors="replace") as err_f:
|
||||
subprocess.run(
|
||||
args,
|
||||
stdin=subprocess.DEVNULL,
|
||||
stdout=subprocess.DEVNULL,
|
||||
stderr=err_f,
|
||||
timeout=120,
|
||||
check=True,
|
||||
)
|
||||
packed_text = Path(tmp_path).read_text()
|
||||
except subprocess.TimeoutExpired:
|
||||
return InternResponse(text="Repomix timed out (120s)", usage={})
|
||||
except subprocess.CalledProcessError:
|
||||
err = Path(err_path).read_text(encoding="utf-8", errors="replace")
|
||||
return InternResponse(text=f"Repomix failed: {err[:500]}", usage={})
|
||||
except FileNotFoundError:
|
||||
return InternResponse(text="npx not found - install Node.js", usage={})
|
||||
finally:
|
||||
# Always clean up the temp files, even on error.
|
||||
for _p in (tmp_path, err_path):
|
||||
try:
|
||||
os.unlink(_p)
|
||||
except OSError:
|
||||
pass
|
||||
|
||||
# Budget check before sending to LLM.
|
||||
tokens = _estimate_tokens(packed_text)
|
||||
if tokens > self.context_budget:
|
||||
return BudgetExceededError(
|
||||
tokens=tokens,
|
||||
limit=self.context_budget,
|
||||
top_files=_top_files_by_size(packed_text),
|
||||
)
|
||||
|
||||
prompt = f"# Packed repo\n\n{packed_text}\n\n# Question\n\n{question}"
|
||||
|
||||
if self.client is None:
|
||||
return InternResponse(text="No client configured", usage={})
|
||||
|
||||
extra_body = kwargs.get("extra_body", {})
|
||||
resp = self.client.chat.completions.create(
|
||||
model=self.model or "deepseek-v4-flash",
|
||||
messages=[
|
||||
{"role": "system", "content": self.system_prompt},
|
||||
{"role": "user", "content": prompt},
|
||||
],
|
||||
max_tokens=kwargs.get("max_tokens", self.max_tokens),
|
||||
temperature=self.temperature,
|
||||
extra_body=extra_body if extra_body else None,
|
||||
)
|
||||
|
||||
choice = resp.choices[0]
|
||||
return InternResponse(
|
||||
text=choice.message.content or "",
|
||||
usage={
|
||||
"tokens_in": resp.usage.prompt_tokens if resp.usage else 0,
|
||||
"tokens_out": resp.usage.completion_tokens if resp.usage else 0,
|
||||
},
|
||||
)
|
||||
46
lib/interns-mcp/interns_mcp/interns/transcript_distill.py
Normal file
46
lib/interns-mcp/interns_mcp/interns/transcript_distill.py
Normal file
@@ -0,0 +1,46 @@
|
||||
"""Transcript distill intern — compress session transcript into structured action-list."""
|
||||
|
||||
from __future__ import annotations
|
||||
|
||||
from pathlib import Path
|
||||
from typing import Any
|
||||
|
||||
from .base import BlockedByPolicy, Intern, InternResponse
|
||||
|
||||
|
||||
class TranscriptDistill(Intern):
|
||||
id = "transcript_distill"
|
||||
description = "Compress a session transcript/log into a structured action-list."
|
||||
|
||||
def run(self, paths: list[str], question: str = "Extract action items, decisions, and unresolved questions.", **kwargs: Any) -> InternResponse | BlockedByPolicy:
|
||||
contents: list[str] = []
|
||||
for p in paths:
|
||||
text = Path(p).read_text(encoding="utf-8", errors="replace")
|
||||
contents.append(f"--- {p} ---\n{text}")
|
||||
|
||||
combined = "\n\n".join(contents)
|
||||
prompt = f"Transcript:\n{combined}\n\nTask: {question}"
|
||||
|
||||
if self.client is None:
|
||||
return InternResponse(text="No client configured", usage={})
|
||||
|
||||
extra_body = kwargs.get("extra_body", {})
|
||||
resp = self.client.chat.completions.create(
|
||||
model=self.model or "deepseek-v4-flash",
|
||||
messages=[
|
||||
{"role": "system", "content": self.system_prompt},
|
||||
{"role": "user", "content": prompt},
|
||||
],
|
||||
max_tokens=kwargs.get("max_tokens", self.max_tokens),
|
||||
temperature=self.temperature,
|
||||
extra_body=extra_body if extra_body else None,
|
||||
)
|
||||
|
||||
choice = resp.choices[0]
|
||||
return InternResponse(
|
||||
text=choice.message.content or "",
|
||||
usage={
|
||||
"tokens_in": resp.usage.prompt_tokens if resp.usage else 0,
|
||||
"tokens_out": resp.usage.completion_tokens if resp.usage else 0,
|
||||
},
|
||||
)
|
||||
Reference in New Issue
Block a user