Adds synapse/slash_commands.py: a chat message that's nothing but
/tool_name(arg=val, arg=val) dispatches straight through tools.dispatch(),
skipping model selection, RAG/playbook context assembly, and the ask-policy
approval round-trip entirely. A human typing this IS the approval - there's
no one else to ask - so it's a deliberate, reviewed bypass of the approval
step specifically, not of anything a tool validates internally (path
boundaries, size caps, Curry's own sandbox checks all still run). Argument
values parse via ast.literal_eval only: strings/numbers/bools/None/literal
containers, no names, no calls, no attribute access - a malformed or
hostile-looking argument fails to parse rather than executing anything.
Wired into chat_stream_endpoint (main.py) as an early short-circuit, before
any of the RAG/model-selection work that a slash-command doesn't need. Web
needed no changes (it already forwards raw text unchanged); the TUI
previously swallowed every leading "/" locally and never reached the backend
with it, so tui_app.py's _handle_slash now falls through to _start_chat for
anything shaped like a tool call while still handling its own local
meta-commands (/help, /model, /new, ...) exactly as before.
Also finally wires Curry in as ten real tools (curry_declare_constant,
curry_get_constant/_latest, curry_list_constants, curry_retire_constant,
curry_declare_function, curry_get_function, curry_list_functions,
curry_call_function, curry_retire_function) - deferred from the vendoring
pass. The five write/execute ones are ACTION tools in the same
always-ask-regardless-of-global-policy floor as edit_source
(ALWAYS_ASK_ACTION_TOOLS, generalized in tools.py from the old
self_edit-only ALWAYS_ASK_TOOLS so future tool families share one place to
register into). curry_call_function is gated as an action for the same
reason run_snippet is: it executes code, even sandboxed.
Fixed a real bug surfaced while wiring this up: curry_db is a long-lived
singleton holding one sqlite3 connection (unlike NexusOS's own memory store,
which opens/closes a fresh connection per call specifically to dodge this),
and sqlite3 forbids using a connection from a different thread than created
it. That's a non-issue in production (uvicorn's single event-loop thread),
but Starlette's TestClient runs the ASGI app through an anyio portal thread,
so it broke immediately under test. Fixed at the source (curry_core.py,
Curry.__init__) with check_same_thread=False, documented as a second
deliberate vendoring deviation alongside the PR #4 sandbox fix - there was
never real concurrent access here, just an overly strict same-thread
assertion tripping on a thread-identity change with only one logical caller.
Verified: 244 backend tests pass (18 new for the parser + endpoint wiring +
curry tool registration, 4 new for the TUI passthrough); the 12 pre-existing
C/C++/Rust toolchain failures are unrelated and unchanged. Confirmed by hand
over the real HTTP endpoint: successful dispatch, zero tool_request events
(approval bypass working as designed), a format()-dunder exploit attempt
still rejected by the vendored sandbox fix even through the new tool
registration, malformed arguments rejected before ever reaching dispatch,
and an unknown tool name rejected cleanly. Wheel rebuilt and content-checked
(bin/check.sh's gate now also asserts slash_commands.py ships).
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Gives the assistant three new ACTION tools to change its own playbooks,
runtime settings, and (source checkout only) its own source code, all
reusing the existing run_snippet/remember approval framework — but with
a hardcoded floor (self_edit.ALWAYS_ASK_TOOLS) so these three always pause
for per-call human approval regardless of the global action_tool_policy
setting. Flipping that policy for an unrelated tool must never silently
also unlock unattended self-modification.
synapse/self_edit.py is the new module doing the actual work, documented
in the same explicit "here's what is and isn't a security boundary" style
as code_run.py:
- edit_source is confined to settings.project_root via the same
realpath + Path.parents boundary check that just closed a sibling-
directory bypass in /icons/image, plus a denylist of dangerous
subtrees (.git, the venv, node_modules, build output, runtime state).
Gated on settings.source_checkout — refuses cleanly in a wheel
install, where there's no live repo to edit or commit into.
- The model sends full file content, never a diff; the server computes
the diff itself via difflib against what's actually on disk, so a
human reviews ground truth, not a description the model wrote.
- Every applied source edit best-effort commits to git as an audit
trail — independent of, not a substitute for, the approval gate.
- edit_playbook merges instead of replacing (main.py's prior
_persist_playbook did a raw replace, which was only safe because the
frontend form always sent a complete object — unsafe for a tool a
model calls with a partial argument set, so this also fixes that
latent bug). Becoming the active system prompt requires an explicit
make_active flag, never a side effect of an ordinary edit.
- edit_settings reuses the existing _SETTINGS_DEFAULTS allowlist.
The approval UI (Chatbot.jsx) previously rendered a tool call's arguments
as Object.values(args).join(", ") in a single-line badge — unusable for
reviewing a diff. It now renders a real, server-computed preview (diff
for source, before/after for playbook/settings) via a new shared
diff-view.js helper, with a loud banner when a change would become the
active system prompt or touch action_tool_policy/system_prompt. A new
nexus-edit fence (self-edit-langs.js + Markdown.jsx's EditBlock) shows
the same diff after an edit is applied, mirroring nexus-run.
Verified: 212 backend tests pass (29 new in test_self_edit.py; the 12
pre-existing C/C++/Rust toolchain failures are unrelated and unchanged),
57 frontend node:test cases pass (15 new), eslint and vite build clean,
and the full approval-preview render path was exercised against the real
built UI with a mocked SSE stream covering all four preview branches
(source diff, playbook becomes-main, settings policy-change, and a
rejected/failing preview).
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
render_preview validates markup and hands it to the browser, which renders it
in an opaque-origin sandboxed iframe. Nothing executes server-side. That model
fits HTML/SVG/JSX and cannot fit C, Rust or Erlang, which need a real
toolchain - so those get a second tool instead of a widened first one.
The split is the feature: the model picks a track by picking a tool, rather
than picking a `lang` value from an enum where half the entries run
server-side and half do not.
synapse/code_run.py compiles and runs one file in a throwaway directory and
returns a ```nexus-run fence carrying the source and its captured output
together, so a model cannot paste output without the code that produced it.
Backticks in the source are re-encoded as ` - still valid JSON, and it
cannot close the fence early.
It is not a sandbox, and the module docstring says so up front. What it gives
is containment by layers: consent (an action tool, gated by
action_tool_policy, per-call Approve/Deny on "ask"), static screening, a
scrubbed environment in a temp dir, wall-clock and POSIX rlimits, and a
network namespace on Linux where unprivileged userns are available. Screening
is a tripwire against a model reaching for `requests` out of habit, not a
boundary against an adversary; layers 1 and 3-5 are the load-bearing ones.
Backend RUN_LANGS and frontend run-langs.js are separate registries because
the two sides need different things - one executes, one labels - and neither
should depend on the other at runtime. tests/test_tools.py asserts the key
sets and the fence tag stay equal, so drift fails the gate instead of
rendering a run result under the wrong language.
tests/snippet_probes/ is a data catalog rather than inlined cases, so adding a
language is a data change and the meta-tests can assert every RUN_LANGS key
has both a smoke probe and a screening probe. Probes skip cleanly on hosts
without the toolchain.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Two tool/agent-layer hardening changes:
* fetch_url now resolves the target host and refuses to connect if any
resolved address is loopback, private (RFC1918/ULA), link-local (incl. the
169.254.169.254 cloud-metadata endpoint), multicast, reserved, or
unspecified. IPv4-mapped IPv6 is unwrapped first, and the guard re-runs on
every redirect hop so a public URL cannot 302 its way to an internal target.
* /chat/approve now requires a single-use token minted when the stream pauses
for approval and delivered only in that stream's tool_request event, compared
in constant time. Previously the pending approval was keyed solely on a
client-supplied conversation_id, so anyone who could enumerate a
conversation_id could approve another client's pending action.
The frontend threads the token from the tool_request event into the approve
call.
Co-authored-by: Cursor <cursoragent@cursor.com>
- Conversations bind to a project on creation; RAG scopes to the
conversation's project, not the global setting.
- Action tools (web_search/fetch_url/remember) are withheld unless
allow_action_tools is enabled (off by default). Settings toggle.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
- Projects/workspaces: documents grouped into projects; chat RAG scopes to the
active project. Switcher in the Documents page.
- Agentic action tools: web_search, fetch_url, and remember (first write tool),
allowlist-gated per playbook.
- Local Whisper STT (faster-whisper, no torch): on-device dictation replacing
the browser Web Speech API. POST /stt + GET /stt/status; browser fallback.
- Vector index extended to conversation recall (message_vectors), with the
brute-force cosine scan kept as the fallback.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
AthenaandClaude Sonnet 5
GitNexus · part of the Nexus stack