The environment section rendered "Model family: Claude Opus 4.6" for
every Anthropic model, so claude-opus-5-5, claude-sonnet-4-6, and any
other Claude model told users it was Opus 4.6.
Render a neutral "Claude" family label followed by the resolved model
ID, and rebuild the system prompt on /model so the line follows a
mid-session switch instead of keeping the startup model.
Merge PR #3360 from c01dsummer/claw-code (fix/roadmap-835-838)
Four fixes addressing issues #3361 and #3360:
1. Parallel tool-result pairing (835): The Anthropic Messages API requires all
tool results of a turn to be in a single message. When a turn issued two tool
calls at once, the results were sent in separate messages, causing a 400 error.
Fixed by merging adjacent tool-result-only messages into one.
2. Terminal-scope path limit (836): Correlation keys up to 512 bytes expanded to
1024+ hex characters, exceeding APFS's 1024-byte total path limit. Fixed by
using SHA256 digest (32 hex chars) for path components while storing the full
key in the record for exact matching.
3. Credential hint (838): When ANTHROPIC_BASE_URL points to a non-Anthropic
endpoint (DeepSeek, corporate gateways), the missing-credentials error now
explains that the endpoint expects its own credentials, not Anthropic's.
4. ROADMAP entries: Documents all three fixes with verification details.
All tests pass (370+ for rusty-claude-cli, 240+ for sdk, 122+ for tools, 209+ for
api). Formatting verified with cargo fmt. CI checks approved and passed.
Fixes#3361, #3360
std::env::vars() panics on any non-UTF-8 value. zsh exports a metafied
$_ when the command path contains non-ASCII characters (e.g. 下載), so
claw panicked on startup. Use vars_os() and skip entries that are not
valid UTF-8.
push_output_block recorded a thinking event at content_block_start (an
empty placeholder) and the stream loop recorded the full block again at
content_block_stop. The empty duplicate was replayed on the next request
and rejected with 400 "each thinking block must contain thinking".
Only emit the event from the non-streaming path.
The /sandbox status can be 'ok' (with namespace isolation) or 'warn' (in
restricted environments like CI without unshare support). Update the test
to accept either value rather than hard-coding an expectation.
- Update client_integration.rs user-agent assertion to expect version 0.1.5
- Fix output_format_contract.rs /sandbox status expectation from 'warn' to 'ok'
The test explicit_model_provider_hint_controls_runtime_provider_dispatch
was failing when OPENAI_BASE_URL was set in the environment to a non-standard
URL (e.g., api.layofflabs.com). The test expects the OpenAI provider to use
the default base URL (api.openai.com) when no explicit base_url is provided
in the ProviderConfig.
This fix adds an EnvVarGuard struct to the tests module and uses it to
ensure OPENAI_BASE_URL is unset before running the test, preventing
environment pollution from affecting the test result.
Fixes: https://github.com/ultraworkers/claw-code/issues/3345
Co-authored-by: clawdbot <clawdbot@ultraworkers.dev>
- Resolve .gitignore conflict: include both nix entries and .claude/
- Update AnthropicRuntimeClient::new() test calls to match new signature
- Fix explicit_model_provider_hint test to unset OPENAI_BASE_URL env var
The PermissionMode enum intentionally does not implement PartialOrd
to prevent silent re-ranking of the privilege ladder. Use the explicit
covers() method for privilege comparisons instead of < and >=.
execute_batch already keyed results to original call slots; this makes
that airtight: panicking worker threads resolve to their own index via
a fallback tuple instead of colliding with slot 0, and results are
asserted to follow call order with a regression test that mixes a
workspace-escape read among valid reads at cap=2.
Adds the fan-out bound: parallel-safe calls run in waves of at most
maxParallelToolCalls (default 8, CLAWD_PARALLEL_TOOL_CALLS env, or the
new setting applied via with_max_parallel_tool_calls at runtime build).
Sequential behavior is preserved: allowed_tools checks still run
before dispatch, and ToolSearch stays off the parallel path.
Review fixes for PR #3250:
1. Restore workspace-level unsafe_code = "forbid" (was downgraded to
"warn" for TUI support). The per-crate [lints.rust] unsafe_code =
"allow" on rusty-claude-cli already overrides it where needed.
2. Fix TaskClaim TOCTOU race in claim_task(). Replace exists()-then-write
with OpenOptions::create_new(true) which atomically fails if the lock
file already exists, eliminating the race window between the check and
the write.
3. Remove TUI scaffolding deps (ratatui, tui-textarea, gag, crossbeam-channel,
libc, unicode-width) and the /tui slash command spec — these belong in
a future TUI PR and are unused in the current source.
Also updates the provider chain precedence test to reflect the new behavior
where the caller's resolved model is the chain primary.
Co-Authored-By: Claw <noreply@openclaw.ai>
Port the agent-team enhancement layer onto upstream/main so the model
can spin up coordinated sub-agent teams for parallel work.
subagentModel config wiring (fix):
- Add subagent_model field to RuntimeFeatureConfig + RuntimeConfig::subagent_model()
accessor so the subagentModel setting (already validated by config_validate.rs)
is now actually read and stored.
- Agent tool's resolve_agent_model falls back to subagentModel from config
when no explicit model is passed.
Provider namespace separation:
- New 'custom-openai' provider kind with dedicated env vars
(CLAWCUSTOMOPENAI_API_KEY / CLAWCUSTOMOPENAI_BASE_URL)
- /setup wizard saves kind: 'custom-openai' for option 5
- Bare model name normalized to 'custom/' prefix to avoid proxy 404s
- Sub-agents inherit /setup-saved provider config via inject_config_as_env_fallbacks
Parallel tool execution:
- Override execute_batch to classify read-only tools as parallel-safe
and run them concurrently via std::thread::scope
- Results return in original model order
Team coordination layer:
- AgentMessage, TaskClaim, TeamStatus tools + shared mailbox directory
- Mode presets (tiny/1x ... mega/6x) + enriched TeamCreate/Agent descriptions
- Background team watcher
- /team slash command: on/off/status/toggle
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
- Add SlashCommand::Tools and SlashCommand::Info enum variants
- Add SLASH_COMMAND_SPECS entries for both commands
- Implement handlers in both REPL and CLI modes
- /tools lists available CLI tools with descriptions
- /info shows claw-code runtime information (version, git sha, rustc, target)
- Both commands return None from handle_slash_command, delegating to REPL/CLI layer
- Add test coverage for both commands in is_none test
- Implementations leverage existing tool specs and binary provenance infrastructure
The terminal check was removed during merge conflict resolution, causing
the spinner animation to be output to non-TTY streams (like pipes used
in tests), which corrupted the output format and broke the
text_prompt_mode_keeps_streamed_text_when_the_turn_fails test.
Restore the is_terminal() check to only animate when stdout is a real
terminal while keeping the interrupt feature.
While a turn runs the main thread is blocked in the conversation loop,
so nothing reads the keyboard. Add an EscapeInterruptMonitor that, for
the duration of a turn on an interactive terminal, switches stdin to a
minimal non-canonical mode (echo/line-buffering off; ISIG and output
processing untouched, unlike full raw mode) and polls for a lone ESC
byte. Escape sequences such as arrow keys are recognized and swallowed.
On Esc it trips the TurnInterruptSignal: the in-flight request is
aborted and control returns to the REPL prompt with the session intact.
Permission prompts read whole lines from stdin mid-turn, so stdin
ownership is coordinated through a StdinPromptGate: while a prompt holds
a lease the listener restores canonical mode and stops consuming bytes.
Also:
- spinner shows '(esc to interrupt)' and the banner/help mention Esc
- WorkerStatus/WorkerEventKind gain an 'interrupted' lifecycle state
- non-terminal stdin and non-unix platforms skip the listener; Ctrl+C
interruption still works there
Closes the Esc-to-interrupt feature request (#3196).
Fixes#3259: When rendering markdown in streaming mode, block-level elements
(code blocks, lists, tables, etc.) were losing trailing newlines due to
render_markdown() calling trim_end(). This caused improper spacing when chunks
were split at block boundaries.
Solution:
- Track last_output_ends_with_newline in MarkdownStreamState to detect when
previous output ended without proper spacing
- In push(): add leading newlines before block elements if previous chunk
didn't end with newlines
- In push(): restore trailing newlines after code block/table closings that
were trimmed by render_markdown()
- Add ensure_newlines() helper to prepare spacing before block starts
- Add comprehensive streaming tests that verify idempotence across all
possible chunk split positions
This ensures that streaming renders produce byte-identical output regardless
of where the input is split into chunks, which is critical for displaying
markdown as it arrives from the API while maintaining proper formatting.
Wire the runtime's TurnInterruptSignal through the CLI turn lifecycle:
- AnthropicRuntimeClient races the streaming request against the
interrupt flag with tokio::select!; when the flag trips, the request
future is dropped, aborting the in-flight HTTP stream for every
provider variant.
- BuiltRuntime installs the shared signal into both the conversation
loop and the API client.
- The per-turn Ctrl+C monitor now interrupts the whole turn instead of
only aborting running hooks, so Ctrl+C mid-turn returns to the REPL
prompt with partial output instead of leaving the request running.
- The REPL spinner reports an interrupted turn distinctly from a
completed one.
Part of the Esc-to-interrupt work (#3196); the Esc key listener lands
separately.
Five defects found by auditing the agent stack, all of which let the harness act
confidently on a wrong picture of its own state.
1. A turn that errored discarded the record of tools that had already run. The CLI
hands run_turn a CLONE of the session; the turn appends the user message, each
assistant reply, and a tool_result per executed tool to that clone alone. The error
path dropped it while the side effects of those tools remained on disk, and the
recovery path then replayed the same user input against a session showing no sign
they had ever run — inviting a second execution of tools that are not idempotent.
The clone is now adopted before teardown, and recovery resumes the turn instead of
replaying it (new ConversationRuntime::resume_turn).
2. Every sizing guard was computed against the model NAME's context window. With a
local server behind a cloud model alias that is 200k against a much smaller
backend, which does not make the preflights inaccurate — it makes them unreachable.
CLAW_CONTEXT_WINDOW now declares the real window; model_token_limit honours it and
returns a limit for unrecognised models instead of None, max_tokens is capped to a
quarter of it, and the auto-compaction threshold derives from it (70%) rather than
defaulting to a 200k-window constant.
3. Whether to recover by discarding conversation history was decided by substring
matching on rendered error prose, including strings that ordinary transport and
decode failures also produce — so a dropped connection cost the session its
transcript. Classification now happens in the API layer, where the error still has
its type and the request is still in hand: typed overflow errors, plus ambiguous
transport failures only when the request actually filled the window. The two
copy-pasted detection blocks are gone.
4. The top-level agent loop had no iteration ceiling (usize::MAX) while spawned
subagents were bounded at 32. It now gets the same bound, overridable with
CLAW_MAX_TURN_ITERATIONS.
5. Auto-continue nudges were injected as USER turns, so harness-authored text was
persisted, re-sent on every later request, and folded into compaction summaries as
something the user had said. They are recorded as system turns now; the wire
payload is unchanged.
Also replaces the repository's own self-host section, which described a translator
and a serving stack no longer in the request path. That file is folded into the
system prompt verbatim, so it was briefing the model with an architecture that does
not exist.
(cherry picked from commit c8834cb4d1a3372eb70241429f641162a420297e)
The suite asserted against whatever the developer's box happened to be
configured as. Two independent leaks, both surfaced by a real `~/.claw`:
1. `parse_args` resolves defaults — notably `permissions.defaultMode` —
through the user-scope config, which `default_config_home()` reads from
`CLAW_CONFIG_HOME` falling back to `$HOME/.claw`. Nothing overrode it, so
a `defaultMode` of `dontAsk` on the host turned 12 assertions from
`WorkspaceWrite` into `DangerFullAccess`. `env_lock()` now returns an
`EnvGuard` that points `CLAW_CONFIG_HOME` at a fresh empty directory and
restores the previous value on drop, so "no user config" is the state
under test. All 42 call sites already bind the guard, so this fixes them
at once; tests that want a populated config home can still set the var
themselves afterwards. `removed_login_and_logout_subcommands_error_helpfully`
asserted a Default-sourced permission mode without taking the guard at
all, and now does.
2. `direct_resume_safe_slash_commands_route_to_local_json_actions_831`
hardcoded `/sandbox` => "warn", which is only true where namespace
isolation is unsupported and just the filesystem sandbox is active
(#731). On a host with working user namespaces the correct answer is
"ok". That entry now accepts either; the test is about #831 routing —
local JSON action instead of `interactive_only` — not about the kernel
under the runner.
Verified with the real ~/.claw/settings.json in place (defaultMode=dontAsk):
cargo test --workspace 0 failures, clippy --workspace --all-targets
-D warnings 0 findings, scripts/fmt.sh clean.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Q9McEEH32ZtyM62GSXVSyJ
(cherry picked from commit 67aa4530c22f6fdf95a29d6a46992af137134c67)
`cargo clippy --workspace --all-targets -- -D warnings` is this repo's own
documented verification command, and it could not run to completion. Earlier
compile failures were masking a further 48 lints; with those cleared the
workspace now passes with zero findings.
Real fixes, not suppressions:
- status_json_value took 8 positional parameters, four of them a tail of
`Option<&_>` that most callers passed as None — trivially transposable.
Grouped into `StatusJsonExtras` with a Default impl; all 7 call sites now
name only the fields they actually set.
- extract_help_metadata's 8-tuple return is now a documented `HelpMetadata`
alias.
- the auto-compact loop indexed preserve_schedule by its loop counter;
it now iterates the array directly.
- a sandbox status chain had two identical arms; merged the condition.
- a filter_map whose every arm returned Some is now a map.
- 33 unused stdout/stderr bindings in the output-format contract tests are
prefixed with `_`, and 9 `map_or(false, ..)` are now `is_some_and(..)`.
NOTE: several `--bin claw` tests (parse_args defaults) read the *real*
`~/.claw/settings.json` rather than an isolated config home, so they fail on
any machine whose user config sets `permissions.defaultMode`. With that file
moved aside the binary's suite is 207/207. That isolation gap is pre-existing
and left for a separate change.
Verified: clippy workspace/all-targets/-D warnings reports 0; cargo test
--workspace passes with the user config neutralised; scripts/fmt.sh clean.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Q9McEEH32ZtyM62GSXVSyJ
(cherry picked from commit 03c8a8fd7ba6e405a8217d00d3766c1c6367e527)
`ApiError` is the `Err` half of nearly every `Result` in the api crate, and
its `Api` variant carried ~139 bytes inline (8 fields, four of them
`Option<String>`). That tripped clippy::result_large_err on 35 signatures
and, with -D warnings, made the whole crate unbuildable under the repo's
own documented verification command.
Move the payload into `ApiErrorDetails` behind a `Box`:
- `ApiError::Api(Box<ApiErrorDetails>)`, with `ApiError::api(..)` to
construct and `api_details()` to borrow
- `ApiErrorDetails` re-exported from the crate root
- adding a field here can no longer re-inflate every signature
This also simplifies `enrich_bearer_auth_error`, which previously
destructured and rebuilt all eight fields at five separate return points
just to append a hint; it now mutates the boxed payload in place.
Also clears the remaining pre-existing lints that blocked -D warnings:
unused test bindings, a redundant `use serde_json`, and an explicit
`.into_iter()` in claw-rag-service.
Verified: cargo test --workspace passes, 0 failures.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Q9McEEH32ZtyM62GSXVSyJ
(cherry picked from commit e9210bd75eb76e5d88b2ae797f537102bcabd41d)
refactor(cli): let the emitter report the output tail it leaves behind
Two places believed they owned terminal output: consume_stream streams
assistant deltas live while emit_output = true, and run_turn decided
whether to emit a separating newline before spinner.finish by guessing
from final_assistant_text(&summary).
That guess agrees with reality only because every *other* writer happens to
terminate its output with a newline — tool results go through
stream_markdown (which appends one), thinking summaries are written with a
trailing newline, and tool-call headers use writeln! — so the assistant
text path is the only one that can leave a partial line. That is a property
of the current renderers rather than of the streaming protocol, and it is
not enforced anywhere.
TerminalTail is state the emitter records into as it writes: both live
writers (AnthropicRuntimeClient::consume_stream and the tool-result blocks
in CliToolExecutor::execute) now push bytes through a TailTrackingWriter
that reports the last byte written, and the caller that ends a turn asks
where the emitter stopped instead of inferring it.
The heuristic also left the failure arm unguarded. A turn that streamed text
and then errored called spinner.fail on a partial line, which the line
clear then erased; that arm now carries the same guard.
Also gate the animated spinner.tick frame on is_terminal(). That frame is
pure cursor control (save, rewrite line, restore) and emits no newline, so
down a pipe it leaves an unterminated banner that the first streamed delta
then continues on the same physical line.
—
*[repo owner's gaebal-gajae (clawdbot) 🦞]*
Route groq-prefixed models through the OpenAI-compatible client,
apply Groq token-limit metadata, and skip local tool injection for
compound models.
Add regression coverage for routing, token limits, and compound
local-tool behavior.
Co-authored-by: Oz <oz-agent@warp.dev>
feat: add persistent auto-memory system
Implement a file-based persistent memory system that allows the AI to retain knowledge about users, feedback, projects, and references across sessions.
## Implementation Details
- MemoryStore data layer with FNV-1a workspace fingerprinting
- MemoryRead/MemoryWrite tools with security hardening (path traversal, symlink, absolute path, and size limit protections)
- System prompt injection of MEMORY.md index with prompt injection mitigation (fenced code block + trust-lowering notice)
- autoMemoryEnabled config option (defaults to true)
- Doctor health check for auto-memory state
- /memory command enhancement showing persistent memory info
## Verification
✓ 1023 lib tests pass
✓ Output format contract tests pass
✓ Release build succeeds
✓ Security hardening verified
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
* fix(runtime): prompt permission mode never invokes the prompter
PermissionMode derives Ord from declaration order, placing Prompt (and Allow)
above the real privilege ladder (ReadOnly < WorkspaceWrite < DangerFullAccess).
In PermissionPolicy::authorize_with_context the ladder check
'current_mode >= required_mode' therefore evaluates true for every tool when the
active mode is Prompt (3 >= {0,1,2}), returning Allow and making the subsequent
'if current_mode == PermissionMode::Prompt' branch that routes to prompt_or_deny
unreachable.
Effect: a session configured with the 'prompt' permission mode silently
auto-allows every tool and never calls the prompter (reached via
ConversationRuntime::run_turn with an interactive prompter), defeating the mode's
entire purpose. The sibling PermissionEnforcer already special-cases Prompt
before its own 'active_mode >= required_mode' check, confirming Prompt is not
meant to be part of the ordered ladder.
Fix: exclude Prompt from the ordered-ladder comparison so Prompt mode falls
through to the interactive prompt path. Allow remains handled by its explicit
equality check. The same comparison appears twice in authorize_with_context —
once in the shared tail and once inside the 'Some(PermissionOverride::Allow)'
arm, which is reached when a hook emits permissionDecision "allow" — so both
copies carry the guard; guarding only the tail still left a hook able to
auto-allow every tool in Prompt mode without ever consulting the prompter.
Adds regression tests covering Prompt mode with a prompter, without a prompter,
and with a hook allow override.
* refactor(runtime): make the permission privilege ladder an explicit rank
`PermissionMode` derived `PartialOrd`/`Ord`, so the ladder deciding whether a
tool runs without asking was declaration order. A variant inserted in the middle
of the enum re-ranked it silently, and the difference between "this tool may
run" and "this tool may run after a prompt" was a property of where an `enum`
line happened to sit. That is how `Prompt` came to outrank `DangerFullAccess`,
satisfy `current_mode >= required_mode` for every tool, and skip the prompter.
Replace the derive with `privilege_rank`, an exhaustive `match` naming every
variant, so inserting one is a build failure rather than a silent re-ranking.
`Ord` is deliberately not derived, so a stray `>=` on two modes no longer
compiles; the two comparisons outside this module now say which rank they read
(`PermissionEnforcer::check_with_required_mode` and the `claw doctor` permission
check). Both keep their previous answers, since `privilege_rank` reproduces the
old declaration order exactly, including `Prompt` and `Allow` above the tiers.
`Prompt` and `Allow` are modes of operation, not privilege tiers, so they are
never ranked *against* a tool requirement. The new `covers` predicate states
that by name rather than by number: `Allow` always covers, `Prompt` never does
(it asks), and only the three tiers compare ranks.
That also collapses the duplicated ladder in `authorize_with_context`. The
`Some(PermissionOverride::Allow)` arm carried a second, identical copy of the
ask-rule check and the ladder; it now falls through to the single copy below,
so a hook `allow` is an override rather than a bypass and the two copies can no
longer drift apart.
Behaviour is unchanged for every (mode, requirement, rule, override)
combination. The tests pin the `Prompt` routes — plain and hook-`allow` — across
all three tiers, plus the rank ordering itself.
Root knowledge base plus complexity-scored subdirectory files for the
rust/ workspace, its five highest-mass crates (runtime, rusty-claude-cli,
api, tools, commands, plugins), and the src/ Python porting workspace.
Generated via init-deep: 13 parallel explore agents, LSP/ast-grep code
map, centrality-scored placement. Snapshot in .omo/init-deep.json (local).
twtrubiks
clawdbot
Bellmanandclawdbot
TheArchitectit
amazd
DaBear987
reacechalmers
黄云龙
AbdullahFageehandOz
Ming LiuandClaude Opus 4.6
gaebal-gajae
YeonGyu-Kim
nankingjing
Aneesh
User