* feat(tools): surface tool failures instead of reporting them as success
A tool can finish without raising and still fail to do what it was asked.
Slack answers HTTP 200 with `{"ok": false, "error": "channel_not_found"}`;
an MCP server sets `isError`; a CrewAI AMP action returns
`API request failed: ...`. In every case the call "worked", so the error
text reached the agent as an ordinary result, the agent narrated the
problem in prose, and the run was recorded as a success.
Concretely: five failed `slackbot_send_message` calls each rendered as
"Tool Execution Completed", the task passed, and the crew passed -- with
the only evidence being a sentence in the final answer. Nothing
downstream could tell the difference, and an agent that keeps going on a
step that silently did nothing builds the rest of its work on it.
Give that outcome a type and a reaction:
- `ToolFailure` -- what a tool returns instead of an error string. The
agent still reads prose via `as_agent_message()`, so model behavior is
unchanged; the framework now knows the call failed.
- `ToolFailurePolicy` -- `ignore` (previous behavior), `warn` (default:
record + emit, keep going), `raise` (abort with
`ToolExecutionFailedError`). Resolved most-specific-first: tool, task,
agent, crew.
- `ToolFailureDetectedEvent` -- emitted before a `raise` aborts, so
subscribers always observe the failure. `ToolUsageFinishedEvent` also
carries a `failure` field so a trace UI can mark the call failed
without correlating two events.
- `tool_failures` on `TaskOutput`, `CrewOutput` and `LiteAgentOutput`,
plus `has_tool_failures`, so consumers never parse a string.
Detection is strictly declarative -- no string sniffing, so a tool that
legitimately returns text about an error is never misread as failing.
Failures come from a returned `ToolFailure`, a raised exception, MCP
`isError`, a spent `max_usage_count`, or an unknown tool.
Wired into all four tool-execution paths (the ReAct path and the three
native function-calling implementations). Sources updated to report
structurally: `MCPClient.call_tool_result()` preserves `isError` that
`call_tool()` dropped, and `CrewAIPlatformActionTool` returns a
`ToolFailure` for non-2xx and for caught exceptions.
Two latent bugs fixed along the way: `ToolUsage` assumed every agent has
a `fingerprint` (LiteAgent does not), and policy resolution now tolerates
malformed values rather than letting telemetry take down a tool call.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01ETacm2dMASfpMAYUiDu5YG
* fix(tools): address review round 1 on tool-failure signalling
Five real defects from Bugbot, none of them cosmetic.
Tool-scoped policy never applied (high). `resolve_tool_failure_policy`
read `tool_failure_policy` off the object handed to it, but every
execution path passes the `CrewStructuredTool` wrapper, which never
carried the attribute -- and `BaseTool` never declared it in the first
place. A tool-scoped `raise`/`ignore` was silently ignored while the
docs and a unit test claimed otherwise; the test passed only because it
called the resolver directly with an authored tool. Declared the field on
`BaseTool`, propagated it through `to_structured_tool()` and
`CrewStructuredTool`, and made resolution fall back through
`_original_tool` so either shape works.
A failed call still printed the green "Completed" panel, then the red
one. That is the terminal version of the exact bug this PR is about.
Suppressed the success panel when the call reported failure.
A raised tool printed twice: `ToolUsageErrorEvent` already renders a red
panel, and the new failure panel repeated it. The event is still emitted
-- policy and traces need it -- but the duplicate console output is gone.
Both decisions now live in named predicates on `ConsoleFormatter` rather
than inline in the listener closure, so they are directly testable.
Unknown tools were reported on the ReAct path but silently ignored on all
three native paths, so the same miss was loud or silent depending on
executor style. Native paths now record `UNKNOWN_TOOL` too. This also
surfaced a live `NameError`: ruff had pruned `ToolFailureReason` from
`agent_utils` as unused, so the new branch would have crashed at runtime.
`LiteAgentOutput` had `tool_failures` but not `has_tool_failures`, which
the PR promised on all three output types -- an `AttributeError` for any
caller sharing one check across result types.
Testing: 16 further tests, 45 total. Two console tests were passing
vacuously because `emit()` dispatches sync handlers on a thread pool, so
the assertions raced the handler; they now assert on the predicates
directly, and the native-path test drains the bus with `flush()` and
checks the synchronously-written record. Full suite still matches
baseline exactly at 377 pre-existing failures.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01ETacm2dMASfpMAYUiDu5YG
* chore: update tool specifications
* fix(tools): address review round 2 and fix CI type failure
CI caught a type error I should have: widening `agent` to accept a
`LiteAgent` (so a standalone LiteAgent resolves its own policy) left the
declared signatures behind. Widened `execute_tool_and_check_finality`,
its async twin, and `ToolCallHookContext` to `Agent | BaseAgent |
LiteAgent | None`, which is what those actually receive now.
Seven CodeRabbit findings, all verified against the code first:
`raise` was being downgraded by three enclosing handlers. With
`max_execution_time` set, `_execute_with_timeout` wrapped every exception
in `RuntimeError`, so `_check_execution_error` no longer recognized the
passthrough and sent the task through the retry loop instead of aborting.
`StepExecutor.execute` turned it into `StepResult(success=False)` and let
the plan continue. `LiteAgent.kickoff` ran it through
`handle_unknown_error` and printed "This is likely a bug - please report
it" for what is a deliberate, configured stop.
Failure records were dropped on two paths. `reset_tool_failures()` only
ran in `_prepare_task_execution`, so `Agent.kickoff()` / `kickoff_async()`
— which enter through `_prepare_kickoff` — accumulated records across
runs. And a guardrail retry calls `execute_task` again, which resets the
agent, so a tool that failed on a blocked attempt vanished from the final
output entirely: a run could report zero failures having demonstrably
failed one. Failures now accumulate across guardrail attempts.
Writing the tests for that surfaced a further miss of my own:
`Agent.kickoff()` builds its `LiteAgentOutput` in `agent/core.py` via
`AgentExecutor`, not through `LiteAgent`, so `tool_failures` was always
empty there regardless of the recording fix. Wired up, and the LiteAgent
path now reads from whichever agent the executor was handed
(`original_agent` under kickoff, `self` standalone) rather than assuming.
`last_tool_failures` returns a copy, so a caller cannot mutate the
agent's record or watch it shift mid-run.
Testing: 7 further tests, 52 total, covering the timeout wrapper, the
retry limit, kickoff reset, the kickoff output path, copy semantics and
guardrail accumulation. Full suite matches baseline exactly at 377
pre-existing failures; mypy clean on every changed file.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01ETacm2dMASfpMAYUiDu5YG
* fix(tools): make crew-scoped policy real and close the last raise leak
Two findings, and the first was a documented feature that never worked.
`resolve_tool_failure_policy` consulted a crew, and the docs advertised
crew as a scope, but `Crew` had no `tool_failure_policy` field at all --
and even with one it was unreachable, because `BaseAgent` defaulted the
policy to `WARN` rather than `None`, so resolution always stopped at the
agent. Crew-level configuration was silently ignored.
Fixed by making "inherit" the default everywhere instead of baking `warn`
into one layer: `Crew` gains the field, and `BaseAgent`/`LiteAgent`
default to `None` like `Task` and `BaseTool` already did. The resolver
owns the single fallback, so the chain is genuinely
tool > task > agent > crew > warn and the effective default with nothing
configured is still `warn`. Reading `agent.tool_failure_policy` now
returns `None` (meaning "inherit") rather than `WARN`.
The other: `StepExecutor` re-raised `ToolExecutionFailedError` from its
outer handler, but the nested handler around the native-to-text tooling
fallback still caught it and returned `StepResult(success=False)`. An
agent whose LLM lacked native tool calling would therefore not abort
under `raise`. That is the third distinct place this exception was being
downgraded; it now re-raises there too.
Testing: 8 further tests, 60 total, including the full precedence chain
walked one level at a time and crew-scoped `raise`/`ignore` driven
end-to-end through `kickoff()` rather than only through the resolver --
the gap that let the original crew bug pass review. Full suite matches
baseline exactly at 377 pre-existing failures; mypy clean on every
changed file.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01ETacm2dMASfpMAYUiDu5YG
* docs: trim comments and docstrings on tool-failure signalling
Prose only -- no behavior change. Cut the module docstring, the longer
class and method docstrings, the multi-line inline comments, and the
verbose Field descriptions down to what actually earns its place. Net 87
lines lighter.
Kept the "why" in every case where the reason is non-obvious (why the
event fires before a raise, why the policy reads through the tool wrapper,
why the bus needs draining in tests) and dropped the restatements of what
the code already says.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01ETacm2dMASfpMAYUiDu5YG
* fix(tools): make ignore truly silent, stop caching failures, close 4 gaps
Six findings from the latest review round, all verified against the code
before touching it.
`ignore` was not silent. `ToolUsageFinishedEvent.failure` was set before
the policy ran, so traces still saw a failed call under a policy documented
as surfacing nothing. Worse, the console then showed *no* panel at all:
green was suppressed because `failure` was present, red was skipped because
`ignore` never emits `ToolFailureDetectedEvent`. New `reportable_failure()`
resolves the policy before the finished event and drops the flag under
`ignore`; wired into all four execution paths.
Failures were being cached. `CacheHandler.add` stored a `ToolFailure` like
any other result, so a transient error became permanent for the rest of the
run and every later hit re-reported a call that never re-ran. The cache now
refuses to store declared failures -- fixed at the single choke point rather
than at each of the four call sites.
A spent `max_usage_count` was invisible on the shared native path.
`BaseTool._claim_usage` returned a bare string that only the executors
recognising that exact message treated as a failure. It now returns a
`ToolFailure` with `USAGE_LIMIT`, so every path records it.
A guardrail returning a whole `TaskOutput` replaced the output without
carrying accumulated failures over, so earlier attempts vanished from
`CrewOutput.tool_failures`. New `merge_tool_failures()` combines and
deduplicates, and the retry-rebuild path uses it too.
A hook-blocked call inherited a cached failure and attributed it to a call
that never ran. Now cleared. Not reachable through the built-in cache once
failures stop being cached, so the test injects a custom cache handler that
does retain them -- verified to fail without the guard.
Also removed a `datetime` import left unused by the earlier console-test
rewrite.
Testing: 13 further tests, 73 total. Full suite matches baseline exactly at
377 pre-existing failures; the usage-limit suites that `_claim_usage`
touches pass unchanged; mypy clean on every changed file.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01ETacm2dMASfpMAYUiDu5YG
* fix(tools): let raise through the parallel native path, guard all handlers
Chasing down CodeRabbit's note about callers of
execute_single_native_tool_call turned up a fifth place this exception was
being downgraded: the experimental executor's parallel branch wrapped
future.result() in a broad except and folded the abort into a fake tool
result, so the remaining parallel calls carried on. The sequential path and
crew_agent_executor's parallel branch were already fine.
Five separate handlers have swallowed this during review, so added a guard
test asserting the passthrough at every site rather than trusting the next
one gets spotted.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01ETacm2dMASfpMAYUiDu5YG
* fix(tools): keep a failed tool out of the final answer, finish crew scope
Three more findings, all confirmed against the code.
A failed `result_as_answer` tool still became the task's output. The native
paths already excluded raised errors and hook blocks from short-circuiting,
but not declared failures -- so an error message silently became the answer,
which is the exact shape of bug this PR exists to prevent. Fixed on all
paths, and there were three independent override points, not one:
`ToolResult.result_as_answer` in tool_utils, the `execution_result`
finality checks in both executors, and `process_tool_results()`, which
reads `agent.tools_results` back separately. The first two fixes alone left
the behavior unchanged; only the third made the test pass.
`ToolUsage` never received a crew, so a crew-level `ignore` half-applied:
recording and `ToolFailureDetectedEvent` stayed quiet, but the flag was
still attached to `ToolUsageFinishedEvent`. It now takes and stores `crew`.
`CrewAgentExecutor.invoke`/`ainvoke` routed a deliberate stop through
`handle_unknown_error`, printing "An unknown error occurred" on verbose
runs. LiteAgent already special-cased this; both now do.
Testing: 5 further tests, 79 total, including that a *successful*
`result_as_answer` tool still short-circuits. Full suite matches baseline
exactly at 377 pre-existing failures; mypy clean on every changed file.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01ETacm2dMASfpMAYUiDu5YG
* fix(tools): report malformed tool args, correlate the failure event
Two findings from the latest round.
Malformed native tool arguments returned early with a plain error dict and
never reported a failure, so `ToolFailureReason.INVALID_INPUT` was declared
but unreferenced -- a bad tool call was absent from records, events and
`raise` aborts. `parse_tool_call_args` now carries an INVALID_INPUT failure
on the error dict and both executors report it before returning.
`ToolFailureDetectedEvent` never set `agent_id`, so a trace could not tie it
to a specific agent instance. Fixing that exposed the same gap running the
other way: `ToolUsage`'s own started/finished/error events never set
`agent_id` either, so on the ReAct path the paired finished event had
nothing to correlate against. Both now set it.
Set explicitly rather than via `from_agent`, which would also overwrite
`agent_role` and lose the `_original_role` preference those events already
apply -- a behavior change that has nothing to do with correlation.
Testing: 5 further tests, 84 total, asserting the ids match between the
failure event and its paired finished event. One existing test pinned the
exact key set of the parse-error dict and was updated for the new key. Full
suite matches baseline at 377 pre-existing failures; the one apparent
addition was the known `test_trace_enable_disable` order-flake, confirmed by
re-running rather than assumed.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01ETacm2dMASfpMAYUiDu5YG
* fix(tools): scope failure accumulation per execution, drop deprecated executor
Two review requests from @lorenzejay.
Accumulation no longer lives as mutable state on the shared agent. A
ContextVar collector is opened around each execution -- task, kickoff, and
each guardrail retry -- and the output reads that collector directly instead
of copying the agent's list. ContextVars are copied per asyncio task and per
thread, so concurrent executions cannot see each other's records, and
nesting is safe for retries. `last_tool_failures` prefers the active
collector and falls back to the last completed execution, so the accessor is
correct during a run too. The per-execution reset that caused the erasure is
gone.
Reproducing this took some digging and the finding is worth recording: crew
tasks *cannot* hit it, because `AgentExecutor` refuses concurrent reuse of
one instance and raises. `agent.kickoff()` has no such guard, and there the
bug reproduces exactly as reported -- two concurrent kickoffs each returned
two records. The regression test forces the overlap with a barrier so it is
deterministic rather than timing-dependent, and I verified it reports [2, 2]
against the old behavior and [1, 1] now.
Removed the tool-failure integration from `CrewAgentExecutor` entirely; that
file is back to its state on main. Note the shared ReAct helper it calls
still records failures, since that is common code rather than new behavior in
the deprecated file -- so a `raise` policy will be swallowed by that
executor's generic handler. Flagged on the PR rather than papered over.
Testing: 89 total. Two tests I wrote for this were vacuous on the first
attempt -- they passed against the simulated pre-fix code -- so each
concurrency test was checked against the old behavior before being kept.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01ETacm2dMASfpMAYUiDu5YG
* fix(tools): report malformed calls everywhere, drop the unused block reason
Four findings.
`execute_single_native_tool_call` swallowed a JSON decode error into an empty
args dict and ran the tool with no input at all -- worse than not reporting
it. It now routes through `parse_tool_call_args` like the executors do, so
the StepExecutor/planning path reports INVALID_INPUT and returns instead of
executing. That also removes a duplicated inline parse.
The ReAct path returned a `ToolUsageError` message as an ordinary result
without reporting it, so a malformed call there was invisible while the
equivalent native failure was recorded. Now reported as INVALID_INPUT too.
`Agent.kickoff` opened a collector but no longer reset the agent-level list,
so `last_tool_failures` grew across kickoffs. Reset restored, matching task
execution.
`ToolFailureReason.BLOCKED_BY_HOOK` was declared and never produced. Rather
than start reporting hook blocks as failures, the member is removed: a block
is a deliberate decision by the hook author, and treating it as a failure
would make `raise` abort on an intentional veto. Added a guard test that every
remaining reason is actually produced somewhere, so a dead member cannot
reappear -- the same smell that flagged INVALID_INPUT last round.
Also switched the deprecation guard test to a single import style.
Testing: 6 further tests, 95 total, including that the tool does not run when
its args fail to parse. Full suite matches baseline at 377 pre-existing
failures; mypy clean.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01ETacm2dMASfpMAYUiDu5YG
* fix(tools): merge failures across kickoff guardrail retries, cancel siblings
Kickoff guardrail retries discarded the blocked attempt's failures. Each
retry calls `_execute_and_build_output`, which opens a fresh collector and
builds a new output, so a run could report zero failures having demonstrably
failed one -- the same bug already fixed on the task guardrail path, which
merges. Now merged there too. Verified the test fails without the fix.
Under `raise`, one parallel native tool aborting left its siblings running:
the pool waited for them and pending ones still started. It now shuts the
pool down with `cancel_futures=True` so a not-yet-started sibling never runs.
Threads already in flight cannot be interrupted in Python, so a concurrent
tool may still complete before the abort surfaces; that is noted at the call
site rather than left implied.
Also satisfied CodeQL by materialising the enum in the guard test's loop.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01ETacm2dMASfpMAYUiDu5YG
---------
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
Co-authored-by: Lorenze Jay <63378463+lorenzejay@users.noreply.github.com>
* fix(tools): sandbox FileWriterTool writes and fix file tool rough edges
FileReadTool confined reads to the working directory, but FileWriterTool
only checked that `filename` stayed inside `directory` — and `directory`
itself is an LLM-supplied schema field. An agent could therefore write
anywhere the process had permission to, including ~/.ssh and site-packages,
while the reader refused to read back what the writer had just written.
FileWriterTool was the only filesystem tool in the package that did not go
through validate_file_path; files_compressor_tool validates even its
output path.
Writes are now confined to base_dir (the working directory by default):
the resolved directory must sit inside base_dir, and the resolved file
must sit inside that directory. The pre-existing filename containment
check is kept as-is and still applies even when the unsafe-paths escape
hatch is on, so no existing guarantee is weakened.
Both tools gain a base_dir field so a developer can widen the sandbox
deliberately instead of reaching for the process-wide
CREWAI_TOOLS_ALLOW_UNSAFE_PATHS kill switch. FileReadTool also stops
rejecting a file_path given to its own constructor: that is
developer-declared intent, and declaring one file does not expose its
siblings.
Also fixed:
- FileReadTool scanned the whole file when reading a line window; it now
stops via islice once the requested lines are collected.
- FileWriterTool._run(**kwargs) made the documented positional call
signature raise TypeError and turned a missing overwrite into
"error accessing key". It now takes named parameters in the documented
(filename, content, directory) order.
- A directory naming an existing file reported "already exists and
overwrite option was not passed" even with overwrite=True; it now
explains the real problem.
- Subdirectories inside filename are created, matching what passing
directory already did.
- Both tools now write and decode UTF-8 by default instead of the
platform locale encoding, with an encoding field to override. The docs
already claimed UTF-8 and recommended the writer to Windows users.
- The writer's schema fields had no descriptions for the LLM.
- Docs claimed FileReadTool parses JSON into a dict (it never has),
shipped a snippet that raised TypeError, and did not mention the path
sandbox. The writer README also began with a stray "Here's the
rewritten README" preamble.
BREAKING CHANGE: FileWriterTool no longer writes outside the working
directory. Pass base_dir to authorize a different tree.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
* chore: update tool specifications
* fix(tools): make the declared FileReadTool file reachable by agents
Addresses review feedback on #6692.
The constructor-path exemption did not actually work the way an agent
calls the tool. The description only advertises a redacted label (the
basename, when the file sits outside the sandbox), but resolution
required the exact absolute path, so the model's call was sandboxed and
the declared file was never read. Worse, file_path was a required schema
field, so the long-documented "call with no arguments to read the default
file" raised a validation error instead:
FileReadTool(file_path="/outside/declared.txt")
.run() -> ValueError: validation failed
.run(file_path="declared.txt") -> Error: File not found
.run(file_path="/outside/declared.txt") -> works, but the model was
never told this path
file_path is now optional in the schema, so omitting it reads the default,
and the declared file is addressable by the label the description shows
the model as well as by its real path. Declaring one file still does not
expose its siblings.
The declared path is also pinned to its real path at construction, so a
later chdir cannot silently repoint it at a different file — previously a
relative constructor path re-resolved against the new working directory
on every call.
Also guards the writer's filepath resolution, which could raise
ValueError out of _run for a filename containing a null byte, breaking
the contract of always returning a descriptive string. The directory and
read paths were already guarded.
Adds docstrings to strtobool and both _run methods, corrects an Arabic
tanween spelling and a kaf-as-descriptor calque in the localized read
docs, and regenerates tool.specs.json for the schema change.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
* fix(tools): anchor the declared read path to base_dir, not the cwd
Addresses the second round of review feedback on #6692.
The previous commit pinned a relative constructor file_path with
os.path.realpath, which anchors to the working directory, while both
format_path_for_display and validate_file_path anchor a relative path to
base_dir. With the two roots disagreeing, the same relative string meant
two different files — and the tool served the cwd one under a label that
looks like it belongs to the sandbox:
FileReadTool(file_path="data.txt", base_dir="/allowed") # cwd=/work
label advertised to the model -> "data.txt"
run(file_path="data.txt") -> contents of /work/data.txt
That reads a file from outside base_dir, so it was a sandbox escape
introduced by the exemption itself, not just a wrong-file bug.
Resolution now goes through a single _resolve_against_base helper that
anchors relative paths exactly the way the sandbox does, so the pinned
path, the advertised label and the containment check all agree. Covered
by test_relative_declared_path_anchors_to_base_dir.
Also softens "always readable" to "always allowed past the containment
check" in the docstring, README and docs, since bypassing containment
does not guarantee the read succeeds — it can still fail on a missing
file, a directory, or permissions.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
* fix(tools): tell the LLM about the path sandbox in tool descriptions
Addresses the low-confidence notes from the Copilot review on #6692.
Both tools' descriptions were pre-sandbox wording, so the model learned
about containment only by attempting a path and reading the error back.
Both now state that access is confined to the tool's allowed directory
and that a path resolving outside it is rejected.
The wording deliberately says "the tool's allowed directory" rather than
"the working directory", because the root is base_dir when one is set,
and naming the absolute root would leak it into the prompt — the same
reason paths are redacted in errors.
Not changed: the notes also suggested advertising `encoding`. That is a
constructor-only field the model cannot set, so describing it to the LLM
would be misleading.
Also fixes a test docstring that contradicted its own assertion — the
public run() path does raise on schema validation failure, which is what
the test asserts.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
* fix(tools): anchor base_dir at construction so the sandbox cannot move
Addresses the third review round on #6692.
Both remaining findings came from the same habit: storing an unanchored
string and re-resolving it later.
A relative base_dir was kept verbatim and re-resolved against getcwd() on
every call, while the declared file was pinned once at construction. After
a chdir the sandbox root moved but the declared default did not, so one
tool applied two different roots. base_dir is now resolved once — in the
reader's __init__, and via a field_validator on the writer so it also
applies on the model_validate path.
That also covers the serialization concern. model_dump drops the private
pin, and __init__ re-runs on restore, so a relative file_path was
re-anchored against whatever the working directory happened to be at load
time. With base_dir anchored, restore rebuilds the identical pin.
The residual case is a relative file_path with no base_dir, where the
sandbox root is the working directory too — so both move together and the
tool stays self-consistent. Covered by
test_declared_path_survives_a_serialization_round_trip and
test_relative_base_dir_is_anchored_at_construction on both tools.
Also corrects the writer's 'directory' description, README and docs: the
default resolves inside the tool's allowed directory, which is base_dir
when one is set, not always the working directory.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
* feat(tools): add WaitTool for pausing on long-running jobs
Agents that kick off out-of-band work (a sandbox build, a deployment, an
async API job) have no way to let clock time pass: they either poll in a
tight loop or give up before the work finishes.
WaitTool pauses for a given number of seconds, with an optional reason
echoed back for traces. A single call waits at most max_seconds (default
300, configurable). Longer requests are clamped to the cap and the result
says so, so the model calls again rather than failing. Sync and async
execution are both implemented; stdlib only, no new dependencies.
The tool description spells out when to reach for it (builds, deploys,
batch jobs, async polling, backoff) and when not to, so models pick it up
for the right reason.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
* fix(tools): enforce non-negative wait on positional calls, fix doc snippets
BaseTool.run() skips args_schema validation when called with positional
arguments, so tool.run(-5) reached time.sleep(-5) and failed with an
unrelated error. _resolve_duration now enforces the seconds >= 0 contract
itself, covered for both run() and arun().
Docs and README examples are now self-contained: check_build_status_tool
is defined with the @tool decorator instead of referenced out of nowhere,
and the async example awaits inside asyncio.run() rather than at top level.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
* docs: point the wait tool card at the edge path
Unprefixed links resolve against the default docs version (v1.15.7),
where the wait tool page does not exist, so the card 404'd in the broken
link check. Prefixing with /edge matches how other edge pages link.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
* fix(tools): never cache waits and keep the advertised cap accurate
Two issues from review, both confirmed against the code.
Waits inherited the default cache_function, which always allows caching.
With crew cache enabled, a repeat call with the same arguments returned
"Waited N seconds." straight from the cache without sleeping, turning a
poll-wait-check loop into a busy loop. WaitTool now declares a
cache_function that always refuses.
The description advertising the cap was only rebuilt when max_seconds
reached __init__ without an explicit description. Passing both (as a
platform building from tool.specs.json init params would), calling
model_validate, or assigning max_seconds left the text claiming 300
seconds while clamping to something else. A model_validator now derives
the description from max_seconds on construction, validation, and
assignment, and leaves a caller-supplied description untouched.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
* fix(tools): reject NaN waits and pluralize single-second results
_resolve_duration now rejects NaN with its own message instead of letting
time.sleep raise "Invalid value NaN (not a number)" from a positional
call. Infinity keeps clamping to the cap like any other oversized wait.
Result and description text no longer says "1 seconds". Tests use the
public WaitTool().description as the baseline rather than reaching for
module-private helpers.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
E2B_API_KEY was declared with required=False on the shared E2B tool
base (E2BExecTool, E2BFileTool, E2BPythonTool), even though none of
the three tools can create or attach to a sandbox without it.
E2B_DOMAIN stays optional since it genuinely defaults to e2b.dev.
Regenerated lib/crewai-tools/tool.specs.json via
generate_tool_specs.py to reflect the change.
bedrock-agentcore 1.7.0 has GHSA-j6g5-3hh3-pgw8 (CVE-2026-16796, high):
argument-delimiter injection in CodeInterpreter.install_packages(). It fails
the pip-audit vulnerability scan on every PR in the repo.
The patch is 1.18.1, which requires boto3>=1.43.31. The old <1.8.0 cap plus
aiobotocore~=3.5.0 (botocore<1.42.92) made that unsatisfiable, so the AWS
stack moves together:
- bedrock-agentcore >=1.7.0,<1.8.0 -> >=1.18.1,<2.0.0
- boto3 ~=1.42.90 -> ~=1.43.46 (aws + bedrock extras)
- aiobotocore ~=3.5.0 -> ~=3.8.0 (aws + bedrock extras)
aiobotocore 3.8.0 allows botocore <1.43.47 and boto3 1.43.46 pins botocore
1.43.46, so the ranges overlap.
Verified: uv lock resolves, pip-audit reports no vulnerabilities (3 existing
ignores, none new), 48 bedrock tests pass, and both bedrock toolkits import
cleanly. BrowserClient.{start,stop,generate_ws_headers} and
CodeInterpreter.{start,stop,invoke} are unchanged in 1.18.1.
* chore: bump json-repair to 0.60.1 and un-ignore fixed vulns in scan
- json-repair 0.25.3 -> 0.60.1 (fixes GHSA-xf7x-x43h-rpqh)
- pyOpenSSL already at 26.2.0 in lock (covers CVE-2026-27448, CVE-2026-27459)
- remove the corresponding --ignore-vuln flags from vulnerability-scan.yml
* fix: adapt _safe_repair_json to json-repair 0.60 semantics
json-repair >= 0.60 returns an empty string for plain-text input and
wraps brace-enclosed junk in a single-element list instead of the old
""/{} sentinel values. Treat both as unrepairable so the original
tool input is preserved.
* chore: fix CI - bump gitpython/pyasn1, drop stale type ignores
- gitpython 3.1.50 -> 3.1.52 (GHSA-2f96-g7mh-g2hx, GHSA-v396-v7q4-x2qj,
GHSA-956x-8gvw-wg5v; fixed in 3.1.51)
- pyasn1 0.6.3 -> 0.6.4 (GHSA-8ppf-4f7h-5ppj, GHSA-hm4w-wwcw-mr6r)
- json-repair 0.60 ships type stubs; remove now-unused
type: ignore[import-untyped] comments flagged by mypy
- added client_name header to the 4 tavily tools to classify incoming requests as 'crewai' requests.
- This is for internal analysis
Co-authored-by: Lorenze Jay <63378463+lorenzejay@users.noreply.github.com>
* docs: add "One Card per Step" Studio page (AGE-107)
Document the merge of the task and agent nodes into a single step card on
the Studio canvas. Written as evergreen present-tense feature docs with a
dated rollout banner (June 24th) for the pre-launch customer announcement;
the banner is the only time-bound content and is flagged for removal after
ship. Added in edge + v1.14.7 across en, pt-BR, ko, and ar, with nav entries
in docs.json and three canvas/editor/swap screenshots.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* fix: bump bedrock agentcore dependencies
---------
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Co-authored-by: alex-clawd <alex@crewai.com>
* feat: update pyproject.toml to specify wheel targets
Added a new section to the pyproject.toml file to include only specific files in the wheel build, enhancing the packaging process. Updated tests to verify the inclusion of these targets.
* feat: add memory save event handling to activity log
Implemented event handlers for MemorySaveStartedEvent, MemorySaveCompletedEvent, and MemorySaveFailedEvent in the crew_run_tui module. This allows the application to log memory save operations, capturing their status and details in the activity log. Added corresponding tests to verify the correct logging behavior for successful and failed memory saves.
* feat: enhance memory save event handling in activity log
Added functionality to suppress nested memory save events and updated the handling of MemorySaveStartedEvent, MemorySaveCompletedEvent, and MemorySaveFailedEvent to improve logging accuracy. Introduced new tests to verify the correct behavior of memory save events, including scenarios for nested events and completion updates for timed-out entries.
* Fix memory save activity log handling
* Normalize alpha package versions
* Update scaffolded crew dependency
* feat: add button to copy setup instructions for CrewAI coding agents
Introduced a button in the documentation that allows users to easily copy setup instructions for CrewAI coding agents. The instructions include installation steps, environment setup, and best practices for using the CrewAI CLI. This enhancement aims to streamline the onboarding process for new users.
* Improve missing CrewAI install guidance
* fix: address pr review feedback
* fix: avoid mismatched memory save rows
* fix: wait for queued memory save events
* fix: avoid matching memory saves on missing ids
* chore: normalize prerelease version to 1.14.8a1
- callable_to_string returns None for lambdas/closures instead of an
unresolvable dotted path; Crew filters Nones out of restored callback
lists.
- EventNode.event serializer honors info.mode so mode='json' calls cascade
properly into nested event payloads.
- RagTool.adapter serializes to None (post-validator rebuilds from
config); concrete adapters hold runtime state that can't be round-tripped.
* feat(tools): declare env_vars on DatabricksQueryTool
Add EnvVar import and env_vars field to DatabricksQueryTool so the host
UI knows which environment variables the tool requires. Both auth paths
(DATABRICKS_HOST+TOKEN or DATABRICKS_CONFIG_PROFILE) are marked
required=False with descriptions explaining the alternative.
* chore: update tool specifications
---------
Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
* fix(tools): correct mongdb typo to pymongo in package_dependencies
The `package_dependencies` field in `MongoDBVectorSearchTool` referenced
the non-existent package `mongdb` instead of the actual PyPI package
`pymongo`, which is the driver imported and used throughout the file.
* chore: update tool specifications
---------
Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
