fix: post-merge review sweep — 8 fixes across security, perf, and correctness by ilblackdragon · Pull Request #1550 · nearai/ironclaw

ilblackdragon · 2026-03-21T21:59:14Z

Summary

Reviewed the past week's merged PRs and closed issues, verified which fixes landed correctly, and addresses 8 residual issues that were either incompletely fixed, closed as false positives but still had code quality gaps, or flagged by automated reviewers on PR #1512.

Issues addressed

[HIGH] Code fence detection logic flawed in extract_suggestions() - uses rfind without #1180 — Code fence detection flawed in extract_suggestions(): rfind("```") couldn't handle odd fence counts (unclosed code blocks). Now counts all fence positions and uses parity to determine inside/outside.
[CRITICAL] **N+1 schema generation on every LLM call** — Tool::schema() in `src/tools/b #1361 (partial) — routine_create_parameters_schema() and event_emit_parameters_schema() were regenerating the full JSON schema on every LLM call. Now cached with OnceLock like their *_discovery_schema() counterparts.
[HIGH] O(n) LRU eviction on every insertion at capacity #1430 — O(n) LRU eviction in embedding cache replaced with lru::LruCache for O(1) insertion, lookup, and eviction.
[CRITICAL] Logic inversion in Telegram auto-verification — `result.verification.is_none() #1281 (code quality) — secret_validated in WASM router now reflects whether any authentication was performed (secret, Ed25519, HMAC), not just whether a secret is configured.
[HIGH] **Prompt injection via unescaped channel/user in lightweight routines** — `bui #1364 (defense-in-depth) — Channel/user fields in routine prompt interpolation are now sanitized (strip newlines, replace backticks, truncate) before injection into LLM prompts.

PR #1512 review comments addressed

Duplicate 401 retry (CRITICAL): send_request() had a 50-line inline 401 retry that conflicted with the outer RetryProvider, causing nested retries with exponential backoff. Removed; now invalidates token and returns RequestFailed (retryable) so RetryProvider handles it.
Token error classification (HIGH): All token exchange errors were mapped to RequestFailed. Now AccessDenied/Expired map to AuthFailed (non-retryable), transient errors remain retryable.
parse_extra_headers() hardcoded key (LOW): Error messages always said LLM_EXTRA_HEADERS regardless of which env var was being parsed. Now accepts the env var name for accurate reporting.

Verified as properly fixed (no action needed)

[HIGH] Unbounded Retry-After duration DoS vulnerability — code accepts any valid u64 #1287 — Retry-After DoS: capped at 1hr, centralized in llm/retry.rs
[HIGH] DRY violation — retry-after parsing logic duplicated across 3 modules (anthrop #1288 — Retry parsing DRY: consolidated across 4 call sites
[HIGH] SSRF risk via configurable embedding base URL without validation #1103 — SSRF via embedding URL: comprehensive validation
[HIGH] Unbounded recursion in validate_tool_schema() #975 — Unbounded recursion: bounded to depth 16
[HIGH] Unbounded parameters schema JSON #977 — Unbounded schema JSON: bounded to 64KB
Routines: full_job runs should stay running until linked job completion #1317, Routines: full_job concurrency limits should track linked job lifetime #1318, Routines: full_job routines should inherit owner approval scope #1316 — Routine lifecycle: all fixed with FullJobWatcher
[CRITICAL] Embedding cache clones embeddings on every hit #1429 — Embedding cache clone overhead: cache-then-move pattern applied

Test plan

cargo clippy --all --benches --tests --examples --all-features — zero warnings
All 42 tests related to changed files pass
New regression test: test_extract_suggestions_inside_unclosed_code_fence
All 11 embedding cache tests pass with lru::LruCache swap
All 39 config/llm tests pass
CI green

🤖 Generated with Claude Code

…rectness 1. Fix code fence detection in extract_suggestions() (issue #1180) - rfind("```") couldn't handle odd fence counts (unclosed blocks) - Now counts all fence positions and checks parity 2. Cache routine parameters_schema() with OnceLock (issue #1361) - routine_create_parameters_schema() and event_emit_parameters_schema() were regenerating JSON on every LLM call 3. Replace O(n) LRU eviction with lru crate (issue #1430) - Embedding cache now uses lru::LruCache for O(1) eviction - Removes manual HashMap + last_accessed tracking 4. Fix WASM router secret_validated semantics (issue #1281) - Now reflects whether any auth (secret/Ed25519/HMAC) was performed - Previously only checked if a secret was configured 5. Sanitize channel/user in routine prompt interpolation (issue #1364) - Defense-in-depth: strip newlines, replace backticks, truncate to 128 chars before injecting into LLM prompt 6. Remove duplicate 401 retry in github_copilot.rs (PR #1512 review) - Internal retry conflicted with outer RetryProvider causing nested retries; now invalidates token and lets RetryProvider handle retry 7. Fix token error classification in github_copilot.rs (PR #1512 review) - AccessDenied/Expired errors now map to AuthFailed (non-retryable) - Transient errors remain RequestFailed (retryable) 8. Fix parse_extra_headers() hardcoded env var name (PR #1512 review) - Error messages now report the actual env var being parsed instead of always saying LLM_EXTRA_HEADERS Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

gemini-code-assist · 2026-03-21T21:59:33Z

Summary of Changes

Hello, I'm Gemini Code Assist¹! I'm currently reviewing this pull request and will post my feedback shortly. In the meantime, here's a summary to help you and other reviewers quickly get up to speed!

This pull request performs a comprehensive post-merge review sweep, integrating several critical fixes and improvements across security, performance, and correctness. It addresses 8 residual issues identified from past PRs and automated reviews, ranging from refining code fence detection and optimizing caching mechanisms to enhancing authentication logic and sanitizing user inputs for LLM prompts. The changes aim to bolster the system's robustness and efficiency by resolving known vulnerabilities and performance bottlenecks.

Highlights

Code Fence Detection: Improved the extract_suggestions() function to correctly handle unclosed code blocks by counting fence positions and using parity to determine if a match is inside or outside a code block, addressing issue [HIGH] Code fence detection logic flawed in extract_suggestions() - uses rfind without #1180.
Schema Regeneration Caching: Introduced OnceLock caching for routine_create_parameters_schema() and event_emit_parameters_schema() to prevent regenerating the full JSON schema on every LLM call, resolving issue [CRITICAL] **N+1 schema generation on every LLM call** — Tool::schema() in `src/tools/b #1361 (partial).
Embedding Cache Performance: Replaced the O(n) LRU eviction mechanism in the embedding cache with lru::LruCache, providing O(1) insertion, lookup, and eviction performance, addressing issue [HIGH] O(n) LRU eviction on every insertion at capacity #1430.
WASM Router Authentication Status: Updated the secret_validated flag in the WASM router to accurately reflect whether any form of authentication (secret, Ed25519, HMAC) was performed, rather than just secret configuration, improving code quality for issue [CRITICAL] Logic inversion in Telegram auto-verification — `result.verification.is_none() #1281.
Prompt Field Sanitization: Implemented sanitization for channel and user fields interpolated into LLM prompts, stripping newlines, replacing backticks, and truncating to prevent prompt injection vulnerabilities, addressing issue [HIGH] **Prompt injection via unescaped channel/user in lightweight routines** — `bui #1364.
Duplicate 401 Retry Removal: Removed the inline 401 retry logic from send_request() in the Copilot provider to avoid conflicts and nested retries with the outer RetryProvider, as identified in PR feat(llm): add GitHub Copilot as LLM provider #1512 review comments.
Token Error Classification: Refined token exchange error handling in the Copilot provider to map AccessDenied and Expired errors to AuthFailed (non-retryable), while transient errors remain RequestFailed (retryable), improving retry logic as per PR feat(llm): add GitHub Copilot as LLM provider #1512 review comments.
Accurate Header Error Reporting: Modified parse_extra_headers() to accept the environment variable name, ensuring error messages accurately report which variable caused the issue instead of always defaulting to LLM_EXTRA_HEADERS, addressing a PR feat(llm): add GitHub Copilot as LLM provider #1512 review comment.

Using Gemini Code Assist

The full guide for Gemini Code Assist can be found on our documentation page, here are some quick tips.

Invoking Gemini

You can request assistance from Gemini at any point by creating a comment using either /gemini <command> or @gemini-code-assist <command>. Below is a summary of the supported commands on the current page.

Feature	Command	Description
Code Review	`/gemini review`	Performs a code review for the current pull request in its current state.
Pull Request Summary	`/gemini summary`	Provides a summary of the current pull request in its current state.
Comment	@gemini-code-assist	Responds in comments when explicitly tagged, both in pull request comments and review comments.
Help	`/gemini help`	Displays a list of available commands.

Customization

To customize Gemini Code Assist for GitHub experience, repository maintainers can create a configuration file and/or provide a custom code review style guide (such as PEP-8 for Python) by creating and adding files to a .gemini/ folder in the base of the repository. Detailed instructions can be found here.

Limitations & Feedback

Gemini Code Assist may make mistakes. Please leave feedback on any instances where its feedback is incorrect or counter productive. You can react with 👍 and 👎 on @gemini-code-assist comments. If you're interested in giving your feedback about your experience with Gemini Code Assist for GitHub and other Google products, sign up here.

Review the Privacy Notices, Generative AI Prohibited Use Policy, Terms of Service, and learn how to configure Gemini Code Assist in GitHub here. Gemini can make mistakes, so double check it and use code with caution. ↩

gemini-code-assist

Code Review

This pull request delivers a valuable collection of fixes across security, performance, and correctness. The changes are well-implemented and address several residual issues effectively. Key improvements include a more robust suggestion extraction logic, caching for schema generation, and a significant performance enhancement to the embedding cache by switching to lru::LruCache. The refactoring of the GitHub Copilot provider to better handle auth errors and retries is also a notable improvement. I have one minor suggestion to improve the performance of the new sanitization function.

gemini-code-assist · 2026-03-21T22:01:08Z

+fn sanitize_prompt_field(value: &str) -> String {
+    const MAX_LEN: usize = 128;
+    let cleaned: String = value
+        .chars()
+        .filter(|c| *c != '\n' && *c != '\r')
+        .take(MAX_LEN)
+        .collect();
+    cleaned.replace('`', "'")


This implementation is correct, but it can be made slightly more efficient. It currently creates an intermediate String via collect(), and then performs another pass over that string with replace(). You can combine these operations into a single pass using map() to handle the backtick replacement during the initial character iteration. This avoids the extra allocation and iteration.

Suggested change

fn sanitize_prompt_field(value: &str) -> String {

const MAX_LEN: usize = 128;

let cleaned: String = value

.chars()

.filter(|c| *c != '\n' && *c != '\r')

.take(MAX_LEN)

.collect();

cleaned.replace('`', "'")

fn sanitize_prompt_field(value: &str) -> String {

const MAX_LEN: usize = 128;

value

.chars()

.filter(|&c| c != '\n' && c != '\r')

.take(MAX_LEN)

.map(|c| if c == '`' { '\'' } else { c })

.collect()

}

References

To improve performance, avoid unnecessary heap allocations. When processing string parts, use iterators directly instead of collecting them into an intermediate data structure like a Vec or String.

Fixed in 8e40f7b — switched to single-pass with .map() as suggested. No intermediate allocation.

Copilot

Pull request overview

Post-merge cleanup PR that finishes/strengthens several previously discussed fixes across correctness (suggestion extraction), performance (schema + embedding caches), and security/defense-in-depth (prompt interpolation + auth handling).

Changes:

Fix <suggestions> extraction to ignore fenced code blocks even with unclosed/odd fence counts; add regression test.
Cache routine/event parameter schemas via OnceLock to avoid regenerating JSON schema per call.
Replace manual embedding-cache LRU eviction with lru::LruCache; adjust GitHub Copilot provider auth error classification and remove nested 401 retry logic; improve extra-header parsing error reporting.

Reviewed changes

Copilot reviewed 7 out of 7 changed files in this pull request and generated 4 comments.

Show a summary per file

File	Description
src/workspace/embedding_cache.rs	Swap manual HashMap+timestamps LRU to `lru::LruCache` implementation.
src/tools/builtin/routine.rs	Cache expensive parameter schema generation with `OnceLock`.
src/llm/github_copilot.rs	Refine token-exchange error classification; remove inline 401 retry and rely on outer RetryProvider.
src/config/llm.rs	Improve extra-header parsing errors by reporting the correct env var name; adjust tests.
src/channels/wasm/router.rs	Adjust `secret_validated` semantics to reflect multiple auth mechanisms.
src/agent/routine_engine.rs	Sanitize notify channel/user fields before interpolating into lightweight routine prompts.
src/agent/dispatcher.rs	Improve code-fence detection for suggestions extraction; add regression test.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

Copilot · 2026-03-21T22:04:21Z

+        // at capacity (O(1)). If another caller already cached this key,
+        // push overwrites it — correct and idempotent.
        {
            let mut guard = self.cache.lock().unwrap_or_else(|e| e.into_inner());


embed() currently always clones and inserts the fetched embedding into the LRU, even if another concurrent caller already populated the same key while the lock was released. That reintroduces extra clone+write work in the thundering-herd case. Re-check the cache under the lock (and just touch/promote the existing entry) before cloning/inserting.

Suggested change

// at capacity (O(1)). If another caller already cached this key,

// push overwrites it — correct and idempotent.

{

let mut guard = self.cache.lock().unwrap_or_else(|e| e.into_inner());

// at capacity (O(1)).

{

let mut guard = self.cache.lock().unwrap_or_else(|e| e.into_inner());

// Re-check cache under the lock in case another concurrent caller

// has already populated this key while the lock was released.

if let Some(existing) = guard.get(&key) {

tracing::trace!("embedding cache filled by concurrent caller");

return Ok(existing.clone());

}

// Still not present; insert our freshly fetched embedding.

Fixed in 8e40f7b — embed() now re-checks cache under lock before cloning. If another concurrent caller already populated the key, we skip the clone and just log a trace.

Copilot · 2026-03-21T22:04:21Z

+        // Cache new embeddings (clone only the cacheable subset), then move
+        // originals into results. LruCache::push handles eviction at capacity.
        {
            let mut guard = self.cache.lock().unwrap_or_else(|e| e.into_inner());
-            let cacheable = miss_indices.len().min(self.config.max_entries);
-            let skip = miss_indices.len() - cacheable;
-            let need_to_evict = (guard.len() + cacheable).saturating_sub(self.config.max_entries);
-            if need_to_evict > 0 {
-                Self::evict_k_oldest(&mut guard, need_to_evict);
-            }
-            let now = Instant::now();
-            for (&orig_idx, emb) in miss_indices[skip..].iter().zip(&new_embeddings[skip..]) {
-                guard.insert(
-                    keys[orig_idx],
-                    CacheEntry {
-                        embedding: emb.clone(),
-                        last_accessed: now,
-                    },
-                );
+            for (&orig_idx, emb) in miss_indices.iter().zip(&new_embeddings) {
+                guard.push(keys[orig_idx], emb.clone());
            }


In embed_batch, the code now clones every newly fetched embedding into the cache via push, even when miss_indices.len() greatly exceeds the cache capacity (those entries will be immediately evicted). This can create large, avoidable clone/CPU overhead for big batches. Consider limiting caching to at most the cache capacity (or a bounded subset) and update the comment that says “clone only the cacheable subset”, which is no longer true.

Fixed in 8e40f7b — embed_batch() now limits caching to the last cap entries via guard.cap().get(), skipping overflow entries that would be immediately evicted.

Copilot · 2026-03-21T22:04:22Z

+    // Call the WASM channel.
+    // If we reach this point, any required secret/signature validation has
+    // already passed (the guards above return 401 on failure). So
+    // `secret_validated` is true whenever a secret was configured and checked.
+    let secret_validated = state.router.requires_secret(channel_name).await
+        || state.router.get_signature_key(channel_name).await.is_some()
+        || state.router.get_hmac_secret(channel_name).await.is_some();


secret_validated is now computed by re-calling requires_secret() / get_signature_key() / get_hmac_secret() even though those values were already fetched earlier for the auth guards. This adds extra awaits and makes the comment misleading (“secret was configured and checked” now includes signature/HMAC too). Prefer threading booleans from the earlier checks (e.g., did_authenticate) and adjust the comment/variable name to reflect “any auth validated”.

Fixed in 8e40f7b — now threads a did_authenticate bool through the auth guards (secret/Ed25519/HMAC) instead of re-calling async methods. Each guard sets it to true on success.

Copilot · 2026-03-21T22:04:22Z

-                return Err(LlmError::AuthFailed {
+                return Err(LlmError::RequestFailed {
                    provider: "github_copilot".to_string(),
+                    reason: format!("HTTP 401: {}", crate::agent::truncate_for_preview(&response_text, 256)),


On HTTP 401, this returns LlmError::RequestFailed with a reason that includes a truncated copy of the response body. RequestFailed’s {reason} is part of the user-facing error string, so this may leak arbitrary backend text. Consider keeping the user-facing reason generic for 401 (status + short message) and only logging the body at TRACE/DEBUG if needed.

Suggested change

reason: format!("HTTP 401: {}", crate::agent::truncate_for_preview(&response_text, 256)),

// Avoid including backend-provided response body in user-facing error.

reason: "HTTP 401 Unauthorized".to_string(),

Fixed in 8e40f7b — 401 error now uses generic "HTTP 401 Unauthorized" message. The response body is already logged at WARN level above (truncated to 256 chars), so it's still available for debugging.

- sanitize_prompt_field: single-pass with map() instead of collect+replace - embed(): re-check cache under lock before cloning (thundering herd) - embed_batch(): limit caching to cache capacity, skip overflow entries - router: thread did_authenticate bool instead of re-calling async methods - github_copilot 401: use generic error message, avoid leaking response body - cargo fmt: fix two formatting violations caught by CI Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

[skip-regression-check] Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

zmanian

Code Review: Post-merge review sweep (8 fixes)

Verdict: APPROVE

All 8 fixes are correct and well-scoped (+114/-170, net reduction).

Per-fix assessment

#	Fix	Verdict
1	Code fence parity detection in `extract_suggestions()`	Sound. Regression test added.
2	`OnceLock` schema caching for routine/event params	Correct, idiomatic.
3	`lru::LruCache` replacing manual O(n) eviction	Sound. Net -73 lines.
4	WASM router `did_authenticate` reflects actual auth	Correct, important for non-secret auth channels.
5	Prompt field sanitization (defense-in-depth)	Good. Single-pass implementation is efficient.
6	Remove duplicate inline 401 retry conflicting with `RetryProvider`	Highest-risk fix, done correctly. Clean separation of concerns.
7	Token error classification (auth non-retryable, transient retryable)	Correct. Aligns with retry/circuit-breaker design.
8	`parse_extra_headers` reports correct env var name	Clean.

Suggestions (non-blocking)

Missing regression tests for fixes 4-7. Per review-discipline.md: "Every bug fix must include a test that would have caught the bug." Consider adding:
- Fix 4: Test channel using Ed25519/HMAC auth verifying secret_validated=true
- Fix 5: Unit test for sanitize_prompt_field (newlines, backticks, truncation)
- Fix 6/7: Test that AccessDenied maps to AuthFailed (non-retryable)
sanitize_prompt_field could strip all control characters (c.is_control()) instead of just \n/\r.
.expect() on NonZeroUsize::new(max_entries) in embedding_cache.rs -- safe due to .max(1) above, but strict reading of project rules says no .expect() in production.

claude · 2026-03-24T01:58:20Z

Code review

Found 6 issues:

[HIGH:92] Quadratic time complexity in fence detection (dispatcher.rs:1106-1110). O(n) per regex match due to take_while() scan. Use binary_search on sorted fence_positions to reduce to O(log n).
[HIGH:85] Missing null-byte validation in sanitize_prompt_field (routine_engine.rs:1311-1319). Filters newlines but not null bytes, which could break downstream parsing. Add c != '\0' to filter.
[MEDIUM:75] Webhook auth uses boolean flag pattern instead of enum (wasm/router.rs:336-516). did_authenticate flag must be set in three places (DRY violation). Use enum AuthMethod instead for exhaustiveness checking.
[MEDIUM:70] Schema caching unnecessarily clones on every access (builtin/routine.rs:607-610). OnceLock forces clone on each call. Use Arc to avoid cloning, or benchmark to justify clone cost.
[MEDIUM:50] Refactored parameter not used as intended (config/llm.rs:455-479). parse_extra_headers_with_key accepts env_var_name but doesn't use it in error messages.
[LOW:50] Off-by-one potential in fence detection (dispatcher.rs:1108). Closure uses fp <= pos to count fences before position. If fence is at exact match position, boundary semantics may be incorrect.

…rectness (nearai#1550) * fix: post-merge review sweep — 8 fixes across security, perf, and correctness 1. Fix code fence detection in extract_suggestions() (issue nearai#1180) - rfind("```") couldn't handle odd fence counts (unclosed blocks) - Now counts all fence positions and checks parity 2. Cache routine parameters_schema() with OnceLock (issue nearai#1361) - routine_create_parameters_schema() and event_emit_parameters_schema() were regenerating JSON on every LLM call 3. Replace O(n) LRU eviction with lru crate (issue nearai#1430) - Embedding cache now uses lru::LruCache for O(1) eviction - Removes manual HashMap + last_accessed tracking 4. Fix WASM router secret_validated semantics (issue nearai#1281) - Now reflects whether any auth (secret/Ed25519/HMAC) was performed - Previously only checked if a secret was configured 5. Sanitize channel/user in routine prompt interpolation (issue nearai#1364) - Defense-in-depth: strip newlines, replace backticks, truncate to 128 chars before injecting into LLM prompt 6. Remove duplicate 401 retry in github_copilot.rs (PR nearai#1512 review) - Internal retry conflicted with outer RetryProvider causing nested retries; now invalidates token and lets RetryProvider handle retry 7. Fix token error classification in github_copilot.rs (PR nearai#1512 review) - AccessDenied/Expired errors now map to AuthFailed (non-retryable) - Transient errors remain RequestFailed (retryable) 8. Fix parse_extra_headers() hardcoded env var name (PR nearai#1512 review) - Error messages now report the actual env var being parsed instead of always saying LLM_EXTRA_HEADERS Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> * fix: address PR review comments and fix formatting - sanitize_prompt_field: single-pass with map() instead of collect+replace - embed(): re-check cache under lock before cloning (thundering herd) - embed_batch(): limit caching to cache capacity, skip overflow entries - router: thread did_authenticate bool instead of re-calling async methods - github_copilot 401: use generic error message, avoid leaking response body - cargo fmt: fix two formatting violations caught by CI Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> * chore: trigger CI re-run with updated refs [skip-regression-check] Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Copilot AI review requested due to automatic review settings March 21, 2026 21:59

Copilot started reviewing on behalf of ilblackdragon March 21, 2026 21:59 View session

gemini-code-assist Bot reviewed Mar 21, 2026

View reviewed changes

Copilot AI reviewed Mar 21, 2026

View reviewed changes

ilblackdragon and others added 2 commits March 21, 2026 15:19

chore: trigger CI re-run with updated refs

3d1a5d3

[skip-regression-check] Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

ilblackdragon requested a review from zmanian March 21, 2026 22:46

github-actions Bot mentioned this pull request Mar 22, 2026

🦞 Bản tin hàng ngày hệ sinh thái OpenClaw 2026-03-22 compasify/agents-radar#71

Open

zmanian approved these changes Mar 23, 2026

View reviewed changes

ilblackdragon merged commit fa51b9f into staging Mar 23, 2026
14 checks passed

ilblackdragon deleted the fix/post-merge-review-sweep branch March 23, 2026 21:50

ironclaw-ci Bot mentioned this pull request Mar 24, 2026

chore: promote staging to staging-promote/dea789cc-23455694329 (2026-03-24 01:54 UTC) #1606

Merged

This was referenced Mar 25, 2026

chore: promote staging to staging-promote/c949521d-23562109203 (2026-03-25 20:47 UTC) #1649

Merged

chore: promote staging to staging-promote/ab0ad948-23563320113 (2026-03-25 21:37 UTC) #1654

Merged

This was referenced Mar 25, 2026

chore: release v0.22.0 #1601

Merged

chore(ironclaw): release v0.23.0 #1658

Merged

github-actions Bot mentioned this pull request Aug 4, 2026

🦞 OpenClaw 生态日报 2026-08-04 96loveslife/big_model_radar#272

Open

-        // at capacity (O(1)). If another caller already cached this key,
-        // push overwrites it — correct and idempotent.
-        {
-            let mut guard = self.cache.lock().unwrap_or_else(|e| e.into_inner());
+        // at capacity (O(1)).
+        {
+            let mut guard = self.cache.lock().unwrap_or_else(|e| e.into_inner());
+            // Re-check cache under the lock in case another concurrent caller
+            // has already populated this key while the lock was released.
+            if let Some(existing) = guard.get(&key) {
+                tracing::trace!("embedding cache filled by concurrent caller");
+                return Ok(existing.clone());
+            }
+            // Still not present; insert our freshly fetched embedding.

	reason: format!("HTTP 401: {}", crate::agent::truncate_for_preview(&response_text, 256)),
	// Avoid including backend-provided response body in user-facing error.
	reason: "HTTP 401 Unauthorized".to_string(),

Conversation

ilblackdragon commented Mar 21, 2026

Summary

Issues addressed

PR #1512 review comments addressed

Verified as properly fixed (no action needed)

Test plan

Uh oh!

gemini-code-assist Bot commented Mar 21, 2026

Summary of Changes

Highlights

Footnotes

Uh oh!

gemini-code-assist Bot left a comment

Choose a reason for hiding this comment

Code Review

Uh oh!

gemini-code-assist Bot Mar 21, 2026

Choose a reason for hiding this comment

Uh oh!

ilblackdragon Mar 21, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Copilot AI Mar 21, 2026

Choose a reason for hiding this comment

Uh oh!

ilblackdragon Mar 21, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Mar 21, 2026

Choose a reason for hiding this comment

Uh oh!

ilblackdragon Mar 21, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Mar 21, 2026

Choose a reason for hiding this comment

Uh oh!

ilblackdragon Mar 21, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Mar 21, 2026

Choose a reason for hiding this comment

Uh oh!

ilblackdragon Mar 21, 2026

Choose a reason for hiding this comment

Uh oh!

zmanian left a comment

Choose a reason for hiding this comment

Code Review: Post-merge review sweep (8 fixes)

Per-fix assessment

Suggestions (non-blocking)

Uh oh!

Uh oh!

claude Bot commented Mar 24, 2026

Code review

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants