Reviewed and verified: 4 new tests added, all passing. 116 lines added to test JSON escape sequences and i64 boundaries. No production code changes.
—
*[repo owner's gaebal-gajae (clawdbot) 🦞]*
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.
validate_workspace_boundary compared paths lexically, so a location inside
the workspace spelled with different letter case (a working directory typed
as c:\users\... for C:\Users\...) or through the \\.\ device namespace was
reported as escaping the workspace.
When the lexical comparison fails, compare the path's canonical form before
reporting an escape. A path whose canonical form is outside the workspace
stays rejected.
accepts_equivalent_extended_path_at_workspace_boundary builds its paths
with backslash separators. Backslashes only separate path components on
Windows, so on Linux each path is a single component, the boundary check
rejects the path, and cargo test fails.
Run the test only on Windows, and call validate_workspace_boundary through
super:: so non-Windows test builds do not warn about an unused import.
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).
Keep extended-length and regular Windows paths comparable when enforcing workspace boundaries, including first-run roots that cannot yet be canonicalized. Add regression coverage for DOS and UNC path forms.
Signed-off-by: Suliman Abdulrazzaq <suliman9000a@gmail.com>
The startup probe validated only the mapping flags against the trivial
program `true`, but the real launcher always adds
--mount --ipc --pid --uts --fork. On environments where the user
namespace is created but mount propagation inside it is restricted
(e.g. AppArmor-restricted CI runners), the fallback mapping passed the
probe and the sandbox activated, yet every sandboxed command died with
"cannot change root filesystem propagation: Permission denied",
silently returning empty tool output and breaking the mock parity
suite.
The candidates now define the complete static launcher shape (mapping
flags + namespace flags), so probe success implies launch success; the
launcher reuses the candidate instead of re-appending the namespace
flags, keeping probe and launch as one source of truth. The
order-guarding test asserts the namespace flags are present in every
candidate.
Co-authored-by: linkst <2024023709@m.scnu.edu.cn>
The fallback candidate relies on the setuid newuidmap/newgidmap helpers
(uidmap package) plus a subuid/subgid range for the current user. Note in
the candidate docs that the startup probe rejects the candidate when those
are missing, so the plain --map-root-user form is used instead.
Plain `unshare --user --map-root-user` fails on kernels and containers
that block unprivileged writes to /proc/self/uid_map (e.g. GitHub Actions,
restricted AppArmor profiles). On those systems util-linux delegates to the
setuid newuidmap/newgidmap helpers when --map-auto is also present.
Add the combined form as a fallback candidate and build the launcher args
from the probed mapping, so systems without newuidmap/newgidmap or a
/etc/subuid range keep using the plain form.
No behavior change. Move the inline probe out of
unshare_user_namespace_works into a reusable unshare_probe helper and a
cached working_unshare_mapping() that picks the first working candidate
from UNSHARE_MAPPING_CANDIDATES, so the launcher and the capability probe
share one code path.
- Update slash_command_specs().len() assertion from 139 to 140.
The /setup command added by this PR increased the spec count by 1
but the test's expected count was not updated, causing CI failure.
- Add assert!(help.contains("/setup")) to the
renders_help_from_shared_specs test so the new command is
verified in the help output.
Fixes CI Build ❌ and Test ❌ on #3218.
The setup wizard was merged in PR #3017 but was orphaned -- it was not
declared as a module in main.rs, making it unreachable. Additionally,
the setup_wizard.rs imports RuntimeProviderConfig which did not exist
on upstream/main. This commit makes the wizard accessible and adds the
necessary RuntimeProviderConfig type.
Changes:
- Add RuntimeProviderConfig struct to runtime/src/config.rs with
kind(), api_key(), base_url(), model() accessors.
- Add parse_optional_provider_config() to parse the provider object
from merged settings JSON.
- Add provider() method to RuntimeConfig and RuntimeFeatureConfig.
- Export RuntimeProviderConfig, save_user_provider_settings,
clear_user_provider_settings, and default_config_home from runtime
crate public API (runtime/src/lib.rs).
- Add mod setup_wizard to rusty-claude-cli/src/main.rs.
- Add claw setup CLI subcommand.
- Add /setup slash command.
- Add Setup variant to SlashCommand enum.
- Add Setup to LocalHelpTopic enum.
- Add setup to diagnostic subcommand matching.
- Add subagentModel to TOP_LEVEL_FIELDS in config_validate.rs.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Session save_to_path now wraps ENOENT errors from rotate and atomic
write with a clear "possible concurrent modification" message instead
of surfacing raw OS errno. Helps operators debugging race conditions
when multiple claw invocations touch the same session file.
Generated with https://github.com/Yeachan-Heo/gajae-code
Co-authored-by: Gajae Code <dev@gajae-code.com>
MCP server config now expands ${VAR} environment variable references
and ~/ home directory prefix in command, args, and url fields. Previously
these values were passed verbatim to execve/URL-parse, causing silent
"No such file or directory" failures for standard config patterns.
Generated with https://github.com/Yeachan-Heo/gajae-code
Co-authored-by: Gajae Code <dev@gajae-code.com>
deep_merge_objects now concatenates arrays when both layers provide
the same key. Previously permissions.allow, hooks.PreToolUse, etc.
from earlier config layers (e.g. ~/.claw/settings.json) were silently
discarded when a later layer (e.g. project .claw/settings.json) set
the same key. Now arrays are merged additively across layers.
Generated with https://github.com/Yeachan-Heo/gajae-code
Co-authored-by: Gajae Code <dev@gajae-code.com>
PermissionRule::parse now normalizes tool_name to lowercase, matching
the runtime convention. Previously "Bash(rm:*)" would never match
because the runtime tool name is lowercase "bash". Same fix applied
to denied_tools list.
Generated with https://github.com/Yeachan-Heo/gajae-code
Co-authored-by: Gajae Code <dev@gajae-code.com>
When input is a prefix of a candidate (e.g., mcp → mcpServers), return
the prefix match directly instead of relying on edit-distance which
would incorrectly suggest env (distance 3) over mcpServers (distance 7).
Generated with https://github.com/Yeachan-Heo/gajae-code
Co-authored-by: Gajae Code <dev@gajae-code.com>
Hook config now supports the Claude Code structured hook format with
partial validation. Invalid hook entries are recorded in invalid_hooks
while valid siblings are retained, following the same pattern as MCP
partial validation (#440).
Key changes:
- RuntimeInvalidHookConfig now includes typed kind field (invalid_hooks_config
or unknown_hook_event) for machine-readable error classification
- Hook parsing collects all invalid entries instead of halting at first error
- Unknown hook event names recorded as invalid without rejecting valid hooks
- Legacy bare-string hooks still load with deprecation warnings
- Claude Code documented format loads without error (matcher + nested hooks)
- config/status/doctor JSON surfaces hook_validation metadata
- classify_error_kind maps hook errors to invalid_hooks_config
Generated with https://github.com/Yeachan-Heo/gajae-code
Co-authored-by: Gajae Code <dev@gajae-code.com>
- Fix latest_session_alias_resolves_most_recent_managed_session test:
the test created sessions with 0 messages, which are now filtered out
by the message_count > 0 check in latest_session_excluding(). Updated
the test to call push_user_text() before saving so sessions have
at least one message and are findable by /resume latest.
- Add distinct error message when all sessions are empty (0 messages).
Previously, the same "no managed sessions found" message was returned
whether there were zero sessions or all sessions had 0 messages. Now:
- No sessions at all → "no managed sessions found in {path}. Start
claw to create a session..."
- Sessions exist but all empty → "all sessions are empty (0 messages)
in {path}. This usually means a fresh claw session is running but
no messages have been sent yet. Wait for a response in your other
session, then try --resume latest again."
- Add test for the all-sessions-empty error path.
Addresses reviewer feedback on #3216.
Three improvements to the /resume command:
1. /resume latest now skips the current empty session
When a new session is created on startup (with 0 messages), /resume
latest previously returned that empty session. Now it skips sessions
with message_count == 0 and excludes the current session ID via the
new exclude_id parameter, so it finds the previous session with
actual conversation history.
2. Unified load_session_excluding() replaces load_session_loose()
The previous load_session_loose() only handled cross-workspace
resume for aliases. The new load_session_excluding() combines the
loose workspace validation logic with the exclude_id parameter,
simplifying the call chain and ensuring all resume paths skip the
current empty session when appropriate.
3. All existing session scanning paths (global root + project-local
.claw/sessions/) are already in place from prior commits, and now
the exclude_id filter is applied consistently across both local
and global session scans.
Changes:
- session_control.rs: Add resolve_reference_excluding() that delegates
from resolve_reference(), adding optional exclude_id filtering for
alias references.
- session_control.rs: Add latest_session_excluding() that delegates
from latest_session(), filtering out excluded session IDs and
sessions with 0 messages in both local and global scan paths.
- session_control.rs: Add load_session_excluding() that replaces
load_session_loose(), combining cross-workspace alias handling with
the exclude_id parameter.
- main.rs: Add load_session_reference_excluding() that delegates from
load_session_reference(), using the new store method.
- main.rs: Wire LiveCli::resume_session() to pass the current session
ID as the exclude_id so /resume latest skips the current empty
session.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Cherry-picked from PR #2816 onto current upstream/main, resolving
conflicts from PR #3015's merge (which added retry_after to ApiError
but some construction sites were missing it).
Commits preserved:
- ade85398: API timeout config, Retry-After header, configurable retry
- TimeoutConfig in HTTP client builder (connect 30s, request 5min)
- CLAW_API_CONNECT_TIMEOUT and CLAW_API_REQUEST_TIMEOUT env vars
- Retry-After header parsing on 429 responses
- ApiTimeoutConfig in runtime config (settings.json)
- 8a883430: retry 400 responses with transient gateway error bodies
- Detects known gateway phrases in 400 response bodies
- Marks them as retryable instead of hard-failing
- ed91a61e: add 'no parseable body' to CONTEXT_WINDOW_ERROR_MARKERS
- Some providers return 400 with 'no parseable body' for oversized
requests instead of a proper context_length_exceeded error
Commits skipped (already in upstream via PR #3015):
- 453ab642: optional id field (already merged)
- baa8d1ba: HTML detection in streaming (already merged)
- 33d2f789: JSON error detection in streaming (already merged)
8 files changed, 299 insertions, 80 deletions
- Add TimeoutConfig to HTTP client builder with connect_timeout (30s)
and request_timeout (5min) defaults, configurable via
CLAW_API_CONNECT_TIMEOUT and CLAW_API_REQUEST_TIMEOUT env vars
- Add with_timeout() builder to both AnthropicClient and
OpenAiCompatClient for per-client timeout configuration
- Parse Retry-After header on 429 responses and use it to override
exponential backoff delay when present
- Add ApiTimeoutConfig to runtime config with apiTimeout settings
in ~/.claw/settings.json (connectTimeout, requestTimeout, maxRetries)
- Add retry_after field to ApiError::Api for propagating rate limit
backoff hints through the retry pipeline
Close two ways the permission system could be bypassed:
- Workspace path traversal: normalize `.`/`..` lexically before the
boundary prefix comparison so paths like `/workspace/../../etc` can no
longer escape the sandbox. Fixed in both the runtime enforcer and the
duplicate check in the tools PowerShell path classifier.
- read-only mode no longer trusts the leading token alone: reject shell
metacharacters (chaining/substitution/redirect/pipe/subshell), drop
interpreters and build drivers (python/node/ruby/cargo/rustc) from the
allow-list, gate `git` to non-mutating subcommands, and reject `find`
actions that execute or delete.
Adds regression tests for both holes. The pre-existing, unrelated
worker_boot git-metadata test failure is not affected by this change.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Add SUPPRESS_CONFIG_WARNINGS_STDERR AtomicBool flag in runtime/config.rs
and expose suppress_config_warnings_for_json_mode() via runtime crate.
In main.rs, scan raw argv for --output-format json before parse_args
and activate the flag so no settings-load warnings reach stderr on any
JSON-mode surface (status, sandbox, system-prompt, mcp list, skills list,
agents list, --resume /config*, etc.).
Text-mode surfaces are unaffected; prose deprecation warnings continue
to appear on stderr.
All 572+ tests pass (one pre-existing worker_boot failure unrelated).
黄云龙
Ming LiuandClaude Opus 4.6
gaebal-gajae
Bellman
Suliman Abdulrazzaq
YeonGyu-Kim
code-yeongyuandlinkst
TheArchitectit
Your NameandClaude Opus 4.8
bellmanandGajae Code
Heo, Sung
Sam LamrabteandClaude Opus 4.8