super/box
1
0
Fork 0
Code Issues 2 Pull Requests Actions Packages Projects Releases Wiki Activity

feat(messaging): balanced TOOL parsing, DM shorthand, box.exec, tools.list

Browse Source 
- response-harvester: extract [TOOL]/[EXEC] JSON args with balanced-brace
  scanning (']' and nesting inside args no longer truncate calls); add
  [DM {...}] shorthand mapping to dm.send; native aliases (dm, box,
  tools) plus arg-synonym normalization; formatters and expanded hints.
- exec-constrained: new read-only box.exec op (27 allowlisted box-ctl
  reads) and tools.list op backed by --list-ops for dynamic discovery.
- prompt_envelope: advertise dm.send/box.exec/tools.list in every timer
  DM; add dm_call builder.
- lookup_engine + regex_patterns.json: canonical tool_call pattern
  accepts the DM engine, ']' in args, one nesting level.
- tests/test_tool_calls.py: 38 tests; docs/INBAND-MESSAGING-SPEC.md:
  accepted decision record (Final).
This commit is contained in:
operator-main
2026-10-06 07:29:16 +00:00
parent 345eb09559
commit f2640397ed
7 changed files with 571 additions and 24 deletions
Download Patch File Download Diff File Expand all files Collapse all files
bin/response-harvester.py
+136 -14
View File
@@ -154,6 +154,12 @@ NATIVE_ALIASES = {
"subagents.spawn": "swarm.spawn",
"subagent.list": "swarm.list",
"subagent.status": "swarm.status",
"dm": "dm.send",
"message.send": "dm.send",
"box": "box.exec",
"box.run": "box.exec",
"tools": "tools.list",
"tools.list": "tools.list",
}
@@ -184,6 +190,23 @@ def normalize_native_call(op, args):
break
if "count" not in args and "n" in args:
args["count"] = args.pop("n")
elif op == "dm.send":
if "message" not in args:
for k in ("text", "body", "content", "msg"):
if k in args:
args["message"] = args.pop(k)
break
if "target" not in args:
for k in ("thread", "sidechat", "channel"):
if k in args:
args["target"] = args.pop(k)
break
elif op == "box.exec":
if "action" not in args:
for k in ("cmd", "verb", "command", "run"):
if k in args:
args["action"] = args.pop(k)
break
elif op.startswith("tmux."):
if "session" not in args:
for k in ("name", "target", "s"):
@@ -198,24 +221,96 @@ def normalize_native_call(op, args):
return op, args
_TOOL_OPEN_RE = re.compile(r"\[(TOOL|EXEC)\s+([a-zA-Z0-9_.-]+)\s*")
_DM_OPEN_RE = re.compile(r"\[DM\s+")
def _extract_balanced_json(s, i):
"""Extract one JSON object starting at s[i] == '{' (brace-aware, string-aware).
Returns (obj, end_index) with end_index just past the closing brace,
or (None, i) when no balanced object is present. Unlike a first-']'
regex this tolerates ']' (and nested objects/arrays) inside args.
"""
if i >= len(s) or s[i] != "{":
return None, i
depth = 0
in_str = False
esc = False
for j in range(i, len(s)):
c = s[j]
if in_str:
if esc:
esc = False
elif c == "\\":
esc = True
elif c == '"':
in_str = False
elif c == '"':
in_str = True
elif c == "{":
depth += 1
elif c == "}":
depth -= 1
if depth == 0:
try:
return json.loads(s[i:j + 1]), j + 1
except Exception:
return None, i
return None, i
def _scan_bracket_calls(text):
"""Yield (op, args) for [TOOL op {...}] / [EXEC op {...}] / [DM {...}].
JSON args are extracted with balanced-brace scanning so ']' inside
strings, arrays, or nested objects no longer truncates the call.
Non-JSON tails keep the legacy first-']' behavior (raw passthrough).
"""
out = []
spans = []
for m in _TOOL_OPEN_RE.finditer(text or ""):
spans.append((m.start(), "tool", m.group(2).strip(), m.end()))
for m in _DM_OPEN_RE.finditer(text or ""):
spans.append((m.start(), "dm", "dm.send", m.end()))
spans.sort()
for _, kind, op, pos in spans:
if pos < len(text) and text[pos] == "{":
args, _ = _extract_balanced_json(text, pos)
if args is None:
continue
if not isinstance(args, dict):
args = {"raw": args}
elif kind == "dm":
continue # [DM ...] requires a JSON object; skip bare forms
else:
end = text.find("]", pos)
if end == -1:
continue
raw_args = text[pos:end].strip()
if not raw_args:
args = {}
else:
try:
args = json.loads(raw_args)
if not isinstance(args, dict):
args = {"raw": args}
except Exception:
args = {"raw": raw_args}
out.append((op, args))
return out
def parse_tool_calls(text):
"""
Extract structured tool/exec calls from assistant messages.
Supports:
1. [TOOL <op> <json_args>] or [EXEC <op> <json_args>]
2. ```box / ```tool / ```exec JSON blocks
1. [TOOL <op> <json_args>] or [EXEC <op> <json_args>] (JSON may nest)
2. [DM <json_args>] shorthand for dm.send
3. ```box / ```tool / ```exec JSON blocks
"""
calls = []
for m in re.finditer(r"\[(?:TOOL|EXEC)\s+([a-zA-Z0-9_.-]+)(?:\s+(.*?))?\]", text or ""):
op = m.group(1).strip()
raw_args = (m.group(2) or "").strip()
args = {}
if raw_args:
try:
args = json.loads(raw_args)
except Exception:
args = {"raw": raw_args}
calls.append((op, args))
calls.extend(_scan_bracket_calls(text))
for m in re.finditer(r"```(?:box|tool|exec)\s*\n(.*?)```", text or "", re.DOTALL):
block = m.group(1).strip()
@@ -311,6 +406,10 @@ def format_tool_result_for_chat(op, raw_output):
data = json.loads(raw_output)
except Exception:
s = str(raw_output).strip()
if op == "box.exec":
if len(s) > 900:
s = s[:900] + "\n…(truncated, refine the call for detail)"
return f"box result:\n```\n{s}\n```"
return s[:500] if len(s) > 500 else s
if op == "health.check" and isinstance(data, dict) and "fleet" in data:
@@ -385,6 +484,26 @@ def format_tool_result_for_chat(op, raw_output):
sw = data.get("swarm", {})
return f"Swarm `{sw.get('swarm_id')}`: {sw.get('status')} ({sw.get('done', 0)}/{sw.get('count', 0)} slots completed)."
if op == "tools.list" and isinstance(data, dict):
ops = data.get("ops", [])
if not ops:
return "No tools registered."
ro = [o["op"] for o in ops if not o.get("side_effecting")]
se = [o["op"] for o in ops if o.get("side_effecting")]
lines = [f"{len(ops)} tools available via [TOOL <op> <args>]."]
lines.append("read-only: " + (", ".join(ro) if ro else "none"))
if se:
lines.append("side-effecting: " + ", ".join(se))
return "\n".join(lines)
if op == "box.exec" and isinstance(data, dict):
if data.get("ok") is False:
return f"box call failed: {data.get('error', 'unknown error')}"
out = json.dumps(data)
if len(out) > 900:
out = out[:900] + "\n…(truncated, refine the call for detail)"
return f"box result:\n```\n{out}\n```"
# General fallback: compact JSON capped to 400 chars
s = json.dumps(data)
return s[:400] + "..." if len(s) > 400 else s
@@ -774,9 +893,12 @@ def process_messages(raw_messages, agent, thread_id, thread_name, last_wm, follo
thread_url = f"https://box.muse-dev.online/thread/{thread_id}"
tool_hint = (
f"[Runtime Context: {thread_url}]\n"
f"Tools available: [TOOL <op> <args>] or curl -sk -X POST https://exec.muse-dev.online/exec\n"
f"Tools: [TOOL <op> <args>] or curl -sk -X POST https://exec.muse-dev.online/exec\n"
f" • [TOOL tools.list {{}}] — discover every op dynamically\n"
f" • [TOOL swarm.spawn {{\"count\": 1, \"task\": \"<task>\"}}] — spawn subagents\n"
f" • [DM {{\"to\": \"<agent>\", \"target\": \"<sidechat>\", \"message\": \"<text>\"}}] — send a DM\n"
f" • [TOOL box.exec {{\"action\": \"fleet-status\"}}] — call box (read-only actions)\n"
f" • [TOOL followup.create {{\"in_m\": 5, \"prompt\": \"<reminder>\"}}]\n"
f" • [TOOL swarm.spawn {{\"count\": 1, \"task\": \"<task>\"}}]\n"
f" • [TOOL health.check {{}}]\n\n"
f"[Directive: Take next action or close with [RESULT <job_id>] <summary>]"
)