# Jingxiao Cai's Blog

Technical blog by Jingxiao Cai about distributed ML runtime systems, backend execution reliability, self-hosted AI agents, debugging, automation, and production engineering.

Canonical site: https://anyech.github.io/jingxiao-cai-blog/
RSS: https://anyech.github.io/jingxiao-cai-blog/feed.xml
Sitemap: https://anyech.github.io/jingxiao-cai-blog/sitemap.xml
Search page: https://anyech.github.io/jingxiao-cai-blog/search.html

## Reading guidance for agents

- This is a static technical blog. Public post URLs are stable top-level `*.html` files.
- Prefer the RSS feed for recent posts and the sitemap for complete URL discovery.
- Pagefind search is fully static and generated from article bodies.
- Posts often discuss AI agents, OpenClaw, distributed ML runtime systems, backend execution reliability, debugging, automation, and operational reliability.
- Treat all post content as public writing, not private operational state.

## Posts

- [The Timeout Was Honest. The Active State Was Not.](https://anyech.github.io/jingxiao-cai-blog/timeout-honest-active-state-was-not.html)
  - Markdown mirror: https://anyech.github.io/jingxiao-cai-blog/timeout-honest-active-state-was-not.html.md
  - Date: 2026-08-25
  - Summary: When an adapter exposes tool items only at completion, a watchdog can see zero active work during a real upstream continuation. Fix the missing lifecycle fact first.
  - Tags: ai-agents, agent-ops, debugging, lifecycles, timeouts, tool-calls
- [A Cheaper Agent Is Not a Cheaper Workflow: Delegate by Workload Class](https://anyech.github.io/jingxiao-cai-blog/cheaper-agent-not-cheaper-workflow-delegate-by-class.html)
  - Markdown mirror: https://anyech.github.io/jingxiao-cai-blog/cheaper-agent-not-cheaper-workflow-delegate-by-class.html.md
  - Date: 2026-08-24
  - Summary: Cheap agent delegation should be promoted by workload class, not model label. Tiny-task savings earned shadow expansion; a larger failure locked its class.
  - Tags: ai-agents, agent-ops, delegation, routing, reliability, cost
- [A Default Is Not Caller Intent: Preserve Provenance in Agent Settings](https://anyech.github.io/jingxiao-cai-blog/default-is-not-caller-intent-agent-setting-provenance.html)
  - Markdown mirror: https://anyech.github.io/jingxiao-cai-blog/default-is-not-caller-intent-agent-setting-provenance.html.md
  - Date: 2026-07-18; updated: 2026-08-24
  - Summary: A resolved value is not enough. Preserve request provenance, selection authority, runtime-reported evidence, constraints, and rejection reasons across adapters.
  - Tags: ai-agents, agent-ops, configuration, routing, provenance, testing
- [A Visible Message Does Not Prove an Agent Wake](https://anyech.github.io/jingxiao-cai-blog/visible-message-does-not-prove-agent-wake.html)
  - Markdown mirror: https://anyech.github.io/jingxiao-cai-blog/visible-message-does-not-prove-agent-wake.html.md
  - Date: 2026-07-14; updated: 2026-08-24
  - Summary: A visible handoff, an observed agent turn, a delivered result, and a broader-workflow continuation are separate claims with separate evidence.
  - Tags: openclaw, ai-agents, agent-ops, multi-agent, reliability, automation
- [Yielded Is Not Finished: Guarding Partial-Clone Git Scans Without Breaking Git](https://anyech.github.io/jingxiao-cai-blog/yielded-is-not-finished-partial-clone-git-guards.html)
  - Markdown mirror: https://anyech.github.io/jingxiao-cai-blog/yielded-is-not-finished-partial-clone-git-guards.html.md
  - Date: 2026-08-23
  - Summary: A yielded Git command can remain alive and materialize missing history. The proportionate fix is narrow classification, explicit process ownership, and independent non-execution proof.
  - Tags: git, ai-agents, reliability, agent-ops, tooling, debugging
- [A Local LLM Router Is Not a Panel Lane Yet](https://anyech.github.io/jingxiao-cai-blog/local-llm-router-not-panel-lane-yet.html)
  - Markdown mirror: https://anyech.github.io/jingxiao-cai-blog/local-llm-router-not-panel-lane-yet.html.md
  - Date: 2026-07-04; updated: 2026-08-23
  - Summary: A private local LLM route can pass health and quality checks and still remain only a candidate until controls, performance, environment integrity, role policy, and activation authority align.
  - Tags: ai-agents, local-llm, openclaw, reliability, agent-ops, tooling
- [LLM Panel Orchestration in OpenClaw: Config-Backed Routing, Timeout Classes, and Honest Dissent Without Chaos](https://anyech.github.io/jingxiao-cai-blog/consult-panel-orchestration-openclaw.html)
  - Markdown mirror: https://anyech.github.io/jingxiao-cai-blog/consult-panel-orchestration-openclaw.html.md
  - Date: 2026-04-03; updated: 2026-08-23
  - Summary: A config-backed OpenClaw review workflow with blind-first evidence, authority-family quorum, source snapshots, artifact-backed completion, bridge-back delivery, and honest degradation.
  - Tags: openclaw, ai-agents, llm, orchestration, devops, multi-model-review
- [Archive Before You Retire: A Transaction Model for Git Worktree Backlogs](https://anyech.github.io/jingxiao-cai-blog/archive-before-retire-git-worktree-backlogs.html)
  - Markdown mirror: https://anyech.github.io/jingxiao-cai-blog/archive-before-retire-git-worktree-backlogs.html.md
  - Date: 2026-08-15
  - Summary: Stale worktrees are not safe to delete just because they look old. Archive-preserve transactions make recovery requirements and protected-tip drift checks explicit.
  - Tags: git, devops, agent-ops, recovery, automation, reliability
- [Thread Affinity Is a Safety Boundary for Agent Work](https://anyech.github.io/jingxiao-cai-blog/thread-affinity-safety-boundary-agent-ops.html)
  - Markdown mirror: https://anyech.github.io/jingxiao-cai-blog/thread-affinity-safety-boundary-agent-ops.html.md
  - Date: 2026-06-24; updated: 2026-08-15
  - Summary: When an agent sees “continue,” exact origin identity separates a same-workstream continuation eligible for normal validation from an unknown, different, or conflicting target.
  - Tags: ai-agents, automation, debugging, openclaw, agent-ops, reliability
- [High Draft Acceptance Does Not Guarantee a Throughput Gain: A Lean CPU Speculative-Decoding Test](https://anyech.github.io/jingxiao-cai-blog/high-draft-acceptance-no-throughput-gain-speculative-decoding.html)
  - Markdown mirror: https://anyech.github.io/jingxiao-cai-blog/high-draft-acceptance-no-throughput-gain-speculative-decoding.html.md
  - Date: 2026-08-12
  - Summary: Across two output lengths, every tested draft depth accepted many proposed tokens but still lost on both runtime-reported generation throughput and request wall time.
  - Tags: llm-inference, speculative-decoding, benchmarking, performance, cpu, measurement
- [Approval Is Not Execution: Deferred Side Effects Without a Second Control Plane](https://anyech.github.io/jingxiao-cai-blog/approval-is-not-execution-deferred-side-effects.html)
  - Markdown mirror: https://anyech.github.io/jingxiao-cai-blog/approval-is-not-execution-deferred-side-effects.html.md
  - Date: 2026-08-11
  - Summary: A deferred-effects manifest can preserve proposed intent, suspend dependency-bound work, and compile a review bundle while native action approval remains the only approval authority for the external effect.
  - Tags: ai-agents, agent-ops, approval-workflows, automation-safety, provenance, audit-logging
- [Declarative Change Propagation: How I Built a Self-Documenting Cron System](https://anyech.github.io/jingxiao-cai-blog/declarative-change-propagation-cron-system.html)
  - Markdown mirror: https://anyech.github.io/jingxiao-cai-blog/declarative-change-propagation-cron-system.html.md
  - Date: 2026-03-27; updated: 2026-08-11
  - Summary: Manifest-driven cron updates keep desired state and documentation aligned. A later update separates active-set equality, retirement by design, and terminal-evidence retention.
  - Tags: devops, automation, cron, infrastructure-as-code, drift-detection, openclaw
- [When Reasoning Eats the Answer: Debugging Empty LLM Completions](https://anyech.github.io/jingxiao-cai-blog/when-reasoning-eats-the-answer-empty-llm-completions.html)
  - Markdown mirror: https://anyech.github.io/jingxiao-cai-blog/when-reasoning-eats-the-answer-empty-llm-completions.html.md
  - Date: 2026-08-06
  - Summary: Empty final content with finish_reason=length can be consistent with reasoning exhausting a shared completion budget. A paired-cap test helps separate that case from model and runtime failure.
  - Tags: llm-debugging, ai-agents, token-budgets, reasoning-models, reliability, self-hosted-ai
- [Building a Pattern Scout That Does Not Chase Its Own Echoes](https://anyech.github.io/jingxiao-cai-blog/building-pattern-scout-does-not-chase-own-echoes.html)
  - Markdown mirror: https://anyech.github.io/jingxiao-cai-blog/building-pattern-scout-does-not-chase-own-echoes.html.md
  - Date: 2026-06-12; updated: 2026-08-06
  - Summary: Pattern scouts need source hygiene, novelty gates, opsec filters, evidence cards, and separate submission metrics so no-candidate does not hide a starved pipeline.
  - Tags: openclaw, ai-agents, workflow, reliability, automation, source-hygiene
- [The Harness Passed. The Claim Did Not: Why Independent Evidence Matters](https://anyech.github.io/jingxiao-cai-blog/harness-passed-claim-did-not-independent-evidence.html)
  - Markdown mirror: https://anyech.github.io/jingxiao-cai-blog/harness-passed-claim-did-not-independent-evidence.html.md
  - Date: 2026-07-17
  - Summary: A green harness is not enough when its “observation” merely repeats the input. Runtime claims need an independent evidence path and a real stop rule.
  - Tags: ai-agents, agent-ops, testing, evidence, review, reliability
- [Multi-Agent Collaboration Needs One Owner per Effect](https://anyech.github.io/jingxiao-cai-blog/multi-agent-collaboration-needs-one-owner-per-effect.html)
  - Markdown mirror: https://anyech.github.io/jingxiao-cai-blog/multi-agent-collaboration-needs-one-owner-per-effect.html.md
  - Date: 2026-07-16
  - Summary: More agents do not create coordination. Reliable collaboration gives each logical request stable identity, each effect one owner, and each request one authoritative terminal state.
  - Tags: ai-agents, multi-agent, coordination, reliability, automation, distributed-systems
- [Stop Reviewing the Review: How Agent Workflows Drift Out of Scope](https://anyech.github.io/jingxiao-cai-blog/stop-reviewing-the-review-agent-scope-creep.html)
  - Markdown mirror: https://anyech.github.io/jingxiao-cai-blog/stop-reviewing-the-review-agent-scope-creep.html.md
  - Date: 2026-07-15
  - Summary: Agent work can drift when every review finding creates another validator, receipt, schema, and review cycle. Freeze the target, classify objections, and budget assurance.
  - Tags: ai-agents, agent-ops, scope-creep, review, reliability, automation
- [An Idea Miner Should Be Allowed to Find Nothing](https://anyech.github.io/jingxiao-cai-blog/idea-miner-allowed-find-nothing.html)
  - Markdown mirror: https://anyech.github.io/jingxiao-cai-blog/idea-miner-allowed-find-nothing.html.md
  - Date: 2026-07-11
  - Summary: The useful output of an idea miner is not a pile of concepts. It is an evidence-qualified advance, hold, or reject decision under a hard research bound.
  - Tags: ai-agents, product-discovery, research, automation, agent-ops, validation
- [A Cold-Start Canary Is Not a Serving SLA](https://anyech.github.io/jingxiao-cai-blog/cold-start-canary-not-serving-sla.html)
  - Markdown mirror: https://anyech.github.io/jingxiao-cai-blog/cold-start-canary-not-serving-sla.html.md
  - Date: 2026-07-10
  - Summary: One successful cold start proves a path. It does not prove concurrent safety, cache independence, recovery, cleanup, or a production serving SLA.
  - Tags: ai-agents, local-llm, agent-ops, reliability, kubernetes, self-hosted-ai
- [Prepared Is Not Authorized: The Activation Packet Pattern for Agent Ops](https://anyech.github.io/jingxiao-cai-blog/prepared-is-not-authorized-agent-activation-packet.html)
  - Markdown mirror: https://anyech.github.io/jingxiao-cai-blog/prepared-is-not-authorized-agent-activation-packet.html.md
  - Date: 2026-07-08
  - Summary: A prepared agent workflow can be useful and still not be authorized. The safe pattern is to turn readiness into an activation packet: scope, evidence, boundary state, rollback limits, specific activation verb, and explicit authority.
  - Tags: ai-agents, agent-ops, automation, reliability, security, openclaw
- [A Credential Boundary Is a Production Feature](https://anyech.github.io/jingxiao-cai-blog/credential-boundary-production-feature-agent-ops.html)
  - Markdown mirror: https://anyech.github.io/jingxiao-cai-blog/credential-boundary-production-feature-agent-ops.html.md
  - Date: 2026-07-07
  - Summary: Do not hand setup credentials to runtime route code just because the next integration step looks obvious. Treat the credential boundary itself as a production feature: narrow identity, bounded token minting, negative proofs, cleanup evidence, and a separate activation gate.
  - Tags: ai-agents, agent-ops, automation, security, reliability, openclaw
- [A Final Summary Needs a Delivery Receipt](https://anyech.github.io/jingxiao-cai-blog/final-summary-needs-delivery-receipt-agent-ops.html)
  - Markdown mirror: https://anyech.github.io/jingxiao-cai-blog/final-summary-needs-delivery-receipt-agent-ops.html.md
  - Date: 2026-07-06
  - Summary: A final summary can exist internally while the user-facing thread still lacks proof of delivery. Treat visible closeout as a receipt-backed side effect, not an implied result of the agent producing text.
  - Tags: ai-agents, agent-ops, automation, reliability, delivery, openclaw
- [An Executor Contract Is Not Production Activation](https://anyech.github.io/jingxiao-cai-blog/executor-contract-not-production-activation.html)
  - Markdown mirror: https://anyech.github.io/jingxiao-cai-blog/executor-contract-not-production-activation.html.md
  - Date: 2026-07-05
  - Summary: A dependency-injected executor contract can document shape, policy, and failure behavior offline. It does not authorize live runner wiring, configuration changes, restarts, or real remote execution.
  - Tags: ai-agents, agent-ops, automation, reliability, security, tooling
- [Before Raising Reindex Concurrency, Prove the Memory Lane](https://anyech.github.io/jingxiao-cai-blog/before-raising-reindex-concurrency-prove-memory-lane.html)
  - Markdown mirror: https://anyech.github.io/jingxiao-cai-blog/before-raising-reindex-concurrency-prove-memory-lane.html.md
  - Date: 2026-07-03
  - Summary: Raising embedding reindex concurrency is not just a performance tweak. It becomes eligible only after endpoint identity, index shape, concurrency path, capacity limits, and final behavioral smoke all line up.
  - Tags: ai-agents, agent-ops, automation, debugging, reliability, tooling
- [A Thread Is Closable When No Local Blocker Remains](https://anyech.github.io/jingxiao-cai-blog/thread-closable-when-no-local-blocker-remains.html)
  - Markdown mirror: https://anyech.github.io/jingxiao-cai-blog/thread-closable-when-no-local-blocker-remains.html.md
  - Date: 2026-07-02
  - Summary: A thread is not closable because it feels quiet. It is closable when the objective and delivery are complete, evidence is checked, no local blocker remains, and residual obligations have accepted owners or factual reopen triggers.
  - Tags: ai-agents, agent-ops, automation, debugging, reliability, tooling
- [Thread Checkpoints Are Not Summaries: Making Agent Work Resume Safely](https://anyech.github.io/jingxiao-cai-blog/thread-checkpoints-agent-ops.html)
  - Markdown mirror: https://anyech.github.io/jingxiao-cai-blog/thread-checkpoints-agent-ops.html.md
  - Date: 2026-06-01; updated: 2026-07-02
  - Summary: A thread checkpoint is not a diary entry. For long-running agent work, it is the compact interface that lets the next session resume safely without replaying the whole conversation or inheriting residual obligations from the wrong lane.
  - Tags: ai-agents, automation, debugging, openclaw, agent-ops, documentation
- [Panel Results Are Not Finished Until You Grade the Panelists](https://anyech.github.io/jingxiao-cai-blog/panel-results-need-panelist-evaluation.html)
  - Markdown mirror: https://anyech.github.io/jingxiao-cai-blog/panel-results-need-panelist-evaluation.html.md
  - Date: 2026-06-29
  - Summary: A panel result is not complete when the reviewers answer. It is complete when the parent workflow evaluates coverage, dissent, degraded lanes, and whether each panelist actually judged the target.
  - Tags: ai-agents, agent-ops, automation, debugging, reliability, tooling
- [Role Split Is Not Model Diversity: Artifact QA Panels for Agent Deliverables](https://anyech.github.io/jingxiao-cai-blog/role-split-artifact-qa-panels-agent-deliverables.html)
  - Markdown mirror: https://anyech.github.io/jingxiao-cai-blog/role-split-artifact-qa-panels-agent-deliverables.html.md
  - Date: 2026-06-27
  - Summary: Artifact QA panels are useful when reviewers hold different jobs: first-time reader, skeptic, acceptance gate, cleanup editor. That is role diversity, not model diversity.
  - Tags: ai-agents, agent-ops, automation, debugging, reliability, tooling
- [Building Fail-Closed Stage Environments for AI Agents on a Small VPS](https://anyech.github.io/jingxiao-cai-blog/fail-closed-stage-environments-ai-agents-vps.html)
  - Markdown mirror: https://anyech.github.io/jingxiao-cai-blog/fail-closed-stage-environments-ai-agents-vps.html.md
  - Date: 2026-04-08; updated: 2026-06-27
  - Summary: An OpenClaw stage-environment pattern for a small VPS: fail-closed testing, zero-production-secret bootstrap, detect-only catalog refresh, and a mock-to-real-to-higher-risk ladder.
  - Tags: openclaw, devops, ai-agents, staging, release-engineering, self-hosted
- [Stop Points Are Deliverables: How Agent Work Helps Prevent Accidental Authorization](https://anyech.github.io/jingxiao-cai-blog/stop-points-are-agent-operations-deliverables.html)
  - Markdown mirror: https://anyech.github.io/jingxiao-cai-blog/stop-points-are-agent-operations-deliverables.html.md
  - Date: 2026-06-26
  - Summary: A good agent closeout does more than say “done.” It records the allowed work that passed, names the boundaries that stayed closed, and identifies the exact next approval gate.
  - Tags: ai-agents, agent-ops, automation, reliability, security, tooling
- [The Old Owner Was Still There: Why Agent Cutovers Need Ownership Proof](https://anyech.github.io/jingxiao-cai-blog/old-owner-agent-cutovers.html)
  - Markdown mirror: https://anyech.github.io/jingxiao-cai-blog/old-owner-agent-cutovers.html.md
  - Date: 2026-06-25
  - Summary: A backend restart exposed an obsolete tunnel that could still reclaim the endpoint. The fix was not just a better restart; it was proving ownership, then putting a stable handoff layer in front of replaceable backends.
  - Tags: ai-agents, automation, debugging, reliability, self-hosted-agents
- [A Canary Is a Boundary, Not a Launch Button](https://anyech.github.io/jingxiao-cai-blog/canary-boundary-not-launch-button-agent-ops.html)
  - Markdown mirror: https://anyech.github.io/jingxiao-cai-blog/canary-boundary-not-launch-button-agent-ops.html.md
  - Date: 2026-06-23
  - Summary: When an agent-run canary passes on a real component, the safest interpretation is not “ship it.” It is “the boundary held; now decide the next gate deliberately.”
  - Tags: ai-agents, agent-ops, reliability, canary-testing, automation, distributed-systems
- [Local Semantic Memory on a Small ARM VPS: How I Got OpenClaw Memory Search Working Without External APIs](https://anyech.github.io/jingxiao-cai-blog/local-semantic-memory-openclaw-arm-vps.html)
  - Markdown mirror: https://anyech.github.io/jingxiao-cai-blog/local-semantic-memory-openclaw-arm-vps.html.md
  - Date: 2026-03-19; updated: 2026-06-23
  - Summary: How local memory search became a broader source-hygiene lesson: direct evidence should outrank generated echoes, and useful recall still needs placement gates.
  - Tags: openclaw, ai-agents, self-hosted, memory, embeddings, devops
- [When Agent Threads Get Too Fat: Treat Context Compaction Like a Reliability Boundary](https://anyech.github.io/jingxiao-cai-blog/agent-threads-context-compaction-reliability-boundary.html)
  - Markdown mirror: https://anyech.github.io/jingxiao-cai-blog/agent-threads-context-compaction-reliability-boundary.html.md
  - Date: 2026-06-22
  - Summary: Context compaction is not just token housekeeping. For long-running agent work, it is a reliability boundary that needs durable checkpoints, scoped continuations, and explicit final-delivery contracts.
  - Tags: openclaw, ai-agents, reliability, workflow, context-engineering
- [Public-Safe Evidence Beats Private Debugging Dumps in Agent PRs](https://anyech.github.io/jingxiao-cai-blog/public-safe-agent-pr-evidence-routing.html)
  - Markdown mirror: https://anyech.github.io/jingxiao-cai-blog/public-safe-agent-pr-evidence-routing.html.md
  - Date: 2026-06-21
  - Summary: When an agent hits a real production-flavored failure, the useful upstream contribution is not a raw log dump. It is a small, public-safe evidence unit that names the failure class, the observed state shape, the proof boundary, and the next verification step.
  - Tags: ai-agents, agent-ops, open-source, debugging, opsec, automation
- [The Screenshot Was Green. The Page Was Wrong: Semantic Validation for Agent Artifacts](https://anyech.github.io/jingxiao-cai-blog/screenshot-green-page-wrong-agent-artifact-validation.html)
  - Markdown mirror: https://anyech.github.io/jingxiao-cai-blog/screenshot-green-page-wrong-agent-artifact-validation.html.md
  - Date: 2026-06-20
  - Summary: A screenshot can have the right size, path, and timestamp while showing the wrong page. Agent artifact checks need semantic validation, not just existence checks.
  - Tags: ai-agents, agent-ops, debugging, automation, reliability, tooling
- [Synthetic Fanout Is Not Production Approval: A Safer Pattern for Agent-Run Distributed Probes](https://anyech.github.io/jingxiao-cai-blog/synthetic-fanout-not-production-approval-agent-probes.html)
  - Markdown mirror: https://anyech.github.io/jingxiao-cai-blog/synthetic-fanout-not-production-approval-agent-probes.html.md
  - Date: 2026-06-19
  - Summary: A bounded synthetic fanout can prove that an agent can coordinate a worker plane. It does not prove data approval, production readiness, or service integration. Keep those gates separate.
  - Tags: ai-agents, agent-ops, distributed-systems, automation, reliability, tooling
- [Patch the Thing You Changed: Why Config Validators Should Stay Narrow](https://anyech.github.io/jingxiao-cai-blog/patch-the-thing-you-changed-narrow-config-validation.html)
  - Markdown mirror: https://anyech.github.io/jingxiao-cai-blog/patch-the-thing-you-changed-narrow-config-validation.html.md
  - Date: 2026-06-18
  - Summary: Config patch tools should not turn a tiny, reversible edit into a full-system interrogation. Validate the touched path, report unrelated drift separately, and keep dry-run evidence distinct from live activation.
  - Tags: ai-agents, agent-ops, configuration, reliability, tooling, openclaw
- [Mock First, Live When Proven: How to Keep Agent Demos Honest](https://anyech.github.io/jingxiao-cai-blog/mock-first-live-when-proven-agent-demos.html)
  - Markdown mirror: https://anyech.github.io/jingxiao-cai-blog/mock-first-live-when-proven-agent-demos.html.md
  - Date: 2026-06-17
  - Summary: Mock paths make agent demos safe to build, but they should never pretend to be live. Keep demo modes explicit, prove the live connector with cheap checks, and fail closed when evidence is missing.
  - Tags: ai-agents, agent-ops, tooling, demo-safety, reliability, openclaw
- [Agent Dispatch Should Be Parent-Owned: Let Routers Produce Contracts, Not Side Effects](https://anyech.github.io/jingxiao-cai-blog/parent-owned-agent-dispatch-router-contracts.html)
  - Markdown mirror: https://anyech.github.io/jingxiao-cai-blog/parent-owned-agent-dispatch-router-contracts.html.md
  - Date: 2026-06-16
  - Summary: Routers can make agent work safer by producing exact-scope dispatch contracts instead of launching workers themselves. The parent workflow should own launch authority, evidence checks, and closeout.
  - Tags: ai-agents, agent-ops, routing, workflow-design, openclaw, reliability
- [Fail-Closing Agent Launches: Why Auth and Readiness Gates Should Block Before Tooling Starts](https://anyech.github.io/jingxiao-cai-blog/fail-closing-agent-launches-auth-readiness-gates.html)
  - Markdown mirror: https://anyech.github.io/jingxiao-cai-blog/fail-closing-agent-launches-auth-readiness-gates.html.md
  - Date: 2026-04-29; updated: 2026-06-16
  - Summary: Why AI-agent tool launches should prove auth intent, isolate ambient credentials, check route readiness, and block before side effects when the launch contract is unhealthy.
  - Tags: ai-agents, security, tooling, reliability, openclaw, auth
- [Container-First Distributed Model Serving: Treat Remote Workers as Disposable Proofs](https://anyech.github.io/jingxiao-cai-blog/container-first-distributed-model-serving-disposable-workers.html)
  - Markdown mirror: https://anyech.github.io/jingxiao-cai-blog/container-first-distributed-model-serving-disposable-workers.html.md
  - Date: 2026-06-15
  - Summary: Remote model-serving workers should start as disposable, contract-tested containers—not permanent host mutations. The useful lesson is how to separate adapter bugs, runtime substrate failures, and production-readiness gates.
  - Tags: ai-agents, distributed-systems, model-serving, openclaw, reliability, agent-ops
- [Handling Gemini Capacity Exhaustion: Fallback Lanes for Reliable Agent Workflows](https://anyech.github.io/jingxiao-cai-blog/gemini-capacity-exhaustion-fallback-lanes.html)
  - Markdown mirror: https://anyech.github.io/jingxiao-cai-blog/gemini-capacity-exhaustion-fallback-lanes.html.md
  - Date: 2026-03-29; updated: 2026-06-15
  - Summary: When a coding-agent route drifts, classify the state first: capacity, auth, upstream wait, passive watch, or local adapter repair.
  - Tags: openclaw, ai-agents, gemini, reliability, devops, llm-ops
- [Ready Is Not a Label: PR Readiness Is a Vector](https://anyech.github.io/jingxiao-cai-blog/ready-is-not-a-label-pr-readiness-vector.html)
  - Markdown mirror: https://anyech.github.io/jingxiao-cai-blog/ready-is-not-a-label-pr-readiness-vector.html.md
  - Date: 2026-06-14
  - Summary: A pull request is not ready because one surface says so. Treat readiness as a vector across code head, CI, mergeability, parser-visible proof, review labels, and maintainer scope appetite.
  - Tags: ai-agents, automation, open-source, pull-requests, openclaw, agent-ops
- [When PR Gates Look Broken, Rebase First](https://anyech.github.io/jingxiao-cai-blog/when-pr-gates-look-broken-rebase-first.html)
  - Markdown mirror: https://anyech.github.io/jingxiao-cai-blog/when-pr-gates-look-broken-rebase-first.html.md
  - Date: 2026-06-03; updated: 2026-06-14
  - Summary: A stale pull request can produce misleading CI and policy failures. Updated with the broader rule that PR readiness is a vector across code, CI, mergeability, proof, review state, and scope fit.
  - Tags: ai-agents, automation, debugging, open-source, openclaw, agent-ops
- [Upgrade Preflight as a Product Habit: Why Target Refreshes Should Be Boring](https://anyech.github.io/jingxiao-cai-blog/upgrade-preflight-product-habit-boring-upgrades.html)
  - Markdown mirror: https://anyech.github.io/jingxiao-cai-blog/upgrade-preflight-product-habit-boring-upgrades.html.md
  - Date: 2026-06-13
  - Summary: A practical upgrade preflight pattern for self-hosted AI agent runtimes: refresh the target, preserve the activation boundary, and make upgrades boring before they are allowed to be exciting.
  - Tags: openclaw, devops, ai-agents, release-engineering, self-hosted
- [When Startup Checks Lie: Rolling Back an OpenClaw Runtime Regression](https://anyech.github.io/jingxiao-cai-blog/openclaw-upgrade-rollback-runtime-regression.html)
  - Markdown mirror: https://anyech.github.io/jingxiao-cai-blog/openclaw-upgrade-rollback-runtime-regression.html.md
  - Date: 2026-04-02; updated: 2026-06-13
  - Summary: A clean OpenClaw upgrade passed startup checks but regressed under real use. This incident report covers the rollback, verifier false alarm, target-refresh follow-up, and upgrade guardrails I kept.
  - Tags: openclaw, devops, ai-agents, incident-response, rollback, reliability
- [Nothing Ran, and That Was the Proof: Negative Smokes for Agent Runtime Offload](https://anyech.github.io/jingxiao-cai-blog/nothing-ran-negative-smoke-agent-runtime-offload.html)
  - Markdown mirror: https://anyech.github.io/jingxiao-cai-blog/nothing-ran-negative-smoke-agent-runtime-offload.html.md
  - Date: 2026-06-11
  - Summary: A post-restart offload smoke passed because the runtime loaded, the safe facades answered, and every unsafe execution path stayed deliberately blocked.
  - Tags: openclaw, ai-agents, automation, runtime, testing, reliability
- [Fallback Is Not Preference: Keep Agent Auth Recovery Out of Session State](https://anyech.github.io/jingxiao-cai-blog/fallback-is-not-preference-agent-auth-recovery.html)
  - Markdown mirror: https://anyech.github.io/jingxiao-cai-blog/fallback-is-not-preference-agent-auth-recovery.html.md
  - Date: 2026-06-10
  - Summary: A recent OpenClaw auth-profile incident showed why source tags matter: automatic fallback can keep the reply alive, but it must not become a sticky user choice.
  - Tags: openclaw, ai-agents, automation, auth, debugging, reliability
- [Do Not Teach the App About the Worker: Status Facades for Agent Runtime Offload](https://anyech.github.io/jingxiao-cai-blog/do-not-teach-app-worker-agent-offload-status.html)
  - Markdown mirror: https://anyech.github.io/jingxiao-cai-blog/do-not-teach-app-worker-agent-offload-status.html.md
  - Date: 2026-06-09
  - Summary: A dry-run offload checkpoint showed why the application layer should see a stable status contract, not worker routes, transport details, or live execution mechanics.
  - Tags: openclaw, ai-agents, automation, runtime, debugging, reliability
- [Reachable Is Not Ready: Capability Matrices for Agent Runtime Offload](https://anyech.github.io/jingxiao-cai-blog/reachable-is-not-ready-agent-runtime-offload.html)
  - Markdown mirror: https://anyech.github.io/jingxiao-cai-blog/reachable-is-not-ready-agent-runtime-offload.html.md
  - Date: 2026-06-08
  - Summary: A read-only offload-runtime probe showed several remote lanes were reachable, but not equivalently ready. The safe next step was a capability matrix and dry-run executor contract, not a queue.
  - Tags: openclaw, ai-agents, automation, runtime, debugging, reliability
- [Append, Don't Rewrite: The Guardrail That Saved My Agent's Memory](https://anyech.github.io/jingxiao-cai-blog/append-only-memory-guardrails-agent-ops.html)
  - Markdown mirror: https://anyech.github.io/jingxiao-cai-blog/append-only-memory-guardrails-agent-ops.html.md
  - Date: 2026-06-07
  - Summary: An assistant accidentally replaced a daily memory note instead of appending to it. A shrink guard caught the damage before publish, and the fix became an append-only rule.
  - Tags: openclaw, ai-agents, automation, debugging, memory, reliability
- [When the Live State Moves: Don't Let Validators Read Yesterday's Store](https://anyech.github.io/jingxiao-cai-blog/when-live-state-moves-agent-validators.html)
  - Markdown mirror: https://anyech.github.io/jingxiao-cai-blog/when-live-state-moves-agent-validators.html.md
  - Date: 2026-06-06
  - Summary: After an OpenClaw storage migration, a validator kept reading the old cron store. The fix was not one patch; it was a source-of-truth rule.
  - Tags: openclaw, ai-agents, automation, debugging, reliability, devops
- [Shadow Indexes Beat Hope: Testing Agent Memory Without Touching Production](https://anyech.github.io/jingxiao-cai-blog/shadow-indexes-agent-memory-without-touching-production.html)
  - Markdown mirror: https://anyech.github.io/jingxiao-cai-blog/shadow-indexes-agent-memory-without-touching-production.html.md
  - Date: 2026-06-05
  - Summary: A shadow memory index let me test session-aware agent recall without touching production. The lesson was simple: prove rebuild cost, latency, and answer quality before changing the live memory lane.
  - Tags: openclaw, ai-agents, memory, retrieval, debugging, reliability
- [Proof the Parser Can Read: Behavior Evidence in Agent PRs](https://anyech.github.io/jingxiao-cai-blog/proof-the-parser-can-read-agent-prs.html)
  - Markdown mirror: https://anyech.github.io/jingxiao-cai-blog/proof-the-parser-can-read-agent-prs.html.md
  - Date: 2026-05-30; updated: 2026-06-03
  - Summary: Human-readable proof is not enough when repository automation enforces a schema. Updated with the boundary that proof override requests are not completion when the real behavior gate is still red.
  - Tags: ai-agents, automation, debugging, open-source, openclaw, agent-ops
- [When Your Tunnel Watchdog Lies: Fixing False SSH Alarms Without Hiding Real Failures](https://anyech.github.io/jingxiao-cai-blog/when-your-tunnel-watchdog-lies.html)
  - Markdown mirror: https://anyech.github.io/jingxiao-cai-blog/when-your-tunnel-watchdog-lies.html.md
  - Date: 2026-06-02
  - Summary: A real transient SSH failure plus a wrapper contract bug turned one tunnel watchdog alert into a lesson: keep degraded alerts visible, but do not label them as monitor crashes.
  - Tags: devops, networking, monitoring, ai-agents, openclaw, reliability
- [Gateway Restart Behavior: What OpenClaw Users Need to Know About Config Changes](https://anyech.github.io/jingxiao-cai-blog/gateway-restart-behavior-openclaw.html)
  - Markdown mirror: https://anyech.github.io/jingxiao-cai-blog/gateway-restart-behavior-openclaw.html.md
  - Date: 2026-03-11; updated: 2026-06-02
  - Summary: Some OpenClaw config changes apply live. Others trigger gateway restarts. Updated with rollback, health-monitor, task-registry, and watchdog false-alarm lessons.
  - Tags: openclaw, devops, ai-agents, configuration, gateway, reliability
- [Freshness Is Not Permission: The Opsec Gate in Agent Blog Pipelines](https://anyech.github.io/jingxiao-cai-blog/freshness-is-not-permission-agent-opsec-gates.html)
  - Markdown mirror: https://anyech.github.io/jingxiao-cai-blog/freshness-is-not-permission-agent-opsec-gates.html.md
  - Date: 2026-05-29
  - Summary: Fresh signals make better writing, but they are not automatic publish permission. A sanitized agent-operations pattern for putting an opsec gate between topic scouts and public posts.
  - Tags: ai-agents, automation, opsec, writing-workflow, openclaw, agent-ops
- [The Monitor Is Not the Contract: Durable Handoffs for Long-Running Agents](https://anyech.github.io/jingxiao-cai-blog/monitor-is-not-contract-agent-handoffs.html)
  - Markdown mirror: https://anyech.github.io/jingxiao-cai-blog/monitor-is-not-contract-agent-handoffs.html.md
  - Date: 2026-05-28
  - Summary: Progress monitors are useful, but they are not the contract. A sanitized agent-operations lesson on why long-running work needs durable handoff artifacts, explicit delivery targets, and a final bridge-back gate.
  - Tags: ai-agents, automation, debugging, reliability, openclaw, agent-ops
- [When Cleanup Reveals the Real Contract: Bridge Gaps, Watchdogs, and Agent Delivery](https://anyech.github.io/jingxiao-cai-blog/cleanup-reveals-agent-delivery-contracts.html)
  - Markdown mirror: https://anyech.github.io/jingxiao-cai-blog/cleanup-reveals-agent-delivery-contracts.html.md
  - Date: 2026-05-27
  - Summary: A sanitized OpenClaw agent-operations pattern: cleanup work exposed the difference between removing stale workflow state and proving that bridge-back, progress watchdog, and final delivery contracts were explicit.
  - Tags: ai-agents, automation, debugging, reliability, openclaw, agent-ops
- [The Checkpoint Is the Interface: Durable Handoffs for Long-Running Agent Work](https://anyech.github.io/jingxiao-cai-blog/checkpoint-is-the-interface-agent-handoffs.html)
  - Markdown mirror: https://anyech.github.io/jingxiao-cai-blog/checkpoint-is-the-interface-agent-handoffs.html.md
  - Date: 2026-05-26
  - Summary: Long-running agent work should not depend on chat memory alone. Treat the checkpoint as the interface: status, result, evidence, owner, and final-delivery target.
  - Tags: ai-agents, automation, reliability, workflow, openclaw, agent-ops
- [When a Dirty-Tree Alert Is Correct: Classify the Artifact Before You Commit](https://anyech.github.io/jingxiao-cai-blog/dirty-tree-alert-review-artifact-agent-ops.html)
  - Markdown mirror: https://anyech.github.io/jingxiao-cai-blog/dirty-tree-alert-review-artifact-agent-ops.html.md
  - Date: 2026-05-25
  - Summary: An automated repository-health alert exposed a boundary problem: classify disposable review scaffolding, retained evidence, and intentional source changes before acting.
  - Tags: ai-agents, git, automation, debugging, reliability, openclaw, agent-ops
- [When a Credential Drift Checker Mistakes a Placeholder for a Secret](https://anyech.github.io/jingxiao-cai-blog/credential-drift-placeholder-agent-ops.html)
  - Markdown mirror: https://anyech.github.io/jingxiao-cai-blog/credential-drift-placeholder-agent-ops.html.md
  - Date: 2026-05-24
  - Summary: A credential drift check flagged an inert placeholder as if it were an active secret. The fix was not to delete compatibility state; it was to teach the checker the difference between present and active.
  - Tags: ai-agents, automation, security, debugging, openclaw, agent-ops
- [When Multiple Agents Share the Same Proof Surface](https://anyech.github.io/jingxiao-cai-blog/multi-agent-proof-surface-coordination.html)
  - Markdown mirror: https://anyech.github.io/jingxiao-cai-blog/multi-agent-proof-surface-coordination.html.md
  - Date: 2026-05-20
  - Summary: Staged proof is only low-risk when shared harnesses, fixtures, queues, and scratch state have owners, namespaces, collision checks, and cleanup rules.
  - Tags: ai-agents, github, pull-requests, proof, staging, coordination, reliability, openclaw, agent-ops
- [When Reviewers Still Ask for Live Proof](https://anyech.github.io/jingxiao-cai-blog/reviewer-demands-live-proof-agent-pr-escalation.html)
  - Markdown mirror: https://anyech.github.io/jingxiao-cai-blog/reviewer-demands-live-proof-agent-pr-escalation.html.md
  - Date: 2026-05-19
  - Summary: A reviewer asking for live proof is not a permission grant; agents need to clarify the live-only concern and route production risk to explicit operator approval.
  - Tags: ai-agents, github, pull-requests, proof, staging, escalation, reliability, openclaw, agent-ops
- [Proof Expires: Why Agent PR Evidence Needs a Refresh Window](https://anyech.github.io/jingxiao-cai-blog/agent-pr-proof-expiration-refresh-window.html)
  - Markdown mirror: https://anyech.github.io/jingxiao-cai-blog/agent-pr-proof-expiration-refresh-window.html.md
  - Date: 2026-05-18
  - Summary: Agent PR proof can be true and stale at the same time; cite old evidence with freshness labels, target revisions, and refresh triggers.
  - Tags: ai-agents, github, pull-requests, proof, staging, reliability, openclaw, agent-ops
- [Proof Without Touching Production: A Safer PR Boundary for Agents](https://anyech.github.io/jingxiao-cai-blog/proof-without-touching-production-agent-pr-boundary.html)
  - Markdown mirror: https://anyech.github.io/jingxiao-cai-blog/proof-without-touching-production-agent-pr-boundary.html.md
  - Date: 2026-05-17
  - Summary: Agent PRs need behavior evidence, but production should not be the default proof surface; use staged harnesses, synthetic state, and honest proof boundaries first.
  - Tags: ai-agents, github, pull-requests, proof, staging, reliability, openclaw, agent-ops
- [When Secrets Become References, Monitors Need Adapters](https://anyech.github.io/jingxiao-cai-blog/structured-secret-reference-monitor-adapters.html)
  - Markdown mirror: https://anyech.github.io/jingxiao-cai-blog/structured-secret-reference-monitor-adapters.html.md
  - Date: 2026-05-16
  - Summary: When credentials move from env strings to structured secret references, standalone monitors need compatibility adapters before they report missing credentials.
  - Tags: ai-agents, openclaw, monitoring, configuration, secrets, reliability, agent-ops
- [Before Opening Another Agent PR, Reduce the Queue First](https://anyech.github.io/jingxiao-cai-blog/before-opening-another-agent-pr-reduce-the-queue.html)
  - Markdown mirror: https://anyech.github.io/jingxiao-cai-blog/before-opening-another-agent-pr-reduce-the-queue.html.md
  - Date: 2026-05-15
  - Summary: Agent PR hygiene starts before the next branch: check upstream, consolidate overlapping fixes, close superseded work with pointers, and keep one review surface.
  - Tags: ai-agents, github, pull-requests, automation, reliability, openclaw, agent-ops
- [Why AI Cron Jobs Need Exact-Exec Drivers Instead of Freeform Agent Prompts](https://anyech.github.io/jingxiao-cai-blog/ai-cron-jobs-exact-exec-drivers.html)
  - Markdown mirror: https://anyech.github.io/jingxiao-cai-blog/ai-cron-jobs-exact-exec-drivers.html.md
  - Date: 2026-04-27; updated: 2026-05-14
  - Summary: AI cron reliability needs deterministic helpers, layered timeout budgets, silent-success monitor semantics, and follow-up checks that can actually observe their targets.
  - Tags: ai-agents, automation, reliability, cron, devops, openclaw
- [When the Reply Exists but the Thread Stayed Silent: An Agent-Ops Visibility Lesson](https://anyech.github.io/jingxiao-cai-blog/when-reply-exists-thread-stayed-silent-agent-ops.html)
  - Markdown mirror: https://anyech.github.io/jingxiao-cai-blog/when-reply-exists-thread-stayed-silent-agent-ops.html.md
  - Date: 2026-05-13
  - Summary: A chat-agent visibility lesson: when final answers stay private, completion needs an explicit visible-send target, bridge-back contract, and duplicate suppression.
  - Tags: ai-agents, discord, automation, reliability, openclaw, agent-ops
- [When a True Alert Is Still the Wrong Page: An Agent-Ops Threshold Lesson](https://anyech.github.io/jingxiao-cai-blog/true-alert-wrong-page-agent-ops.html)
  - Markdown mirror: https://anyech.github.io/jingxiao-cai-blog/true-alert-wrong-page-agent-ops.html.md
  - Date: 2026-05-11
  - Summary: A technically true row-count alert became an alert-tuning lesson: record weak proxy crossings, but page only when they combine with real pressure.
  - Tags: ai-agents, automation, alerting, reliability, openclaw, agent-ops
- [When a Coding-Agent Route Drifts: Closing the Loop Without Premature Fixes](https://anyech.github.io/jingxiao-cai-blog/coding-agent-route-drift-without-premature-fixes.html)
  - Markdown mirror: https://anyech.github.io/jingxiao-cai-blog/coding-agent-route-drift-without-premature-fixes.html.md
  - Date: 2026-05-10
  - Summary: A degraded coding-agent lane is not automatically a local repair task; first classify the state as passive watch, upstream wait, or a narrow adapter fix.
  - Tags: ai-agents, coding-agents, openclaw, gemini, reliability, devops
- [When the Report Exists but Delivery Failed: An Agent-Ops Triage Pattern](https://anyech.github.io/jingxiao-cai-blog/when-report-exists-but-delivery-failed-agent-ops.html)
  - Markdown mirror: https://anyech.github.io/jingxiao-cai-blog/when-report-exists-but-delivery-failed-agent-ops.html.md
  - Date: 2026-05-09
  - Summary: A daily scan job generated its report, but the final delivery side effect failed; the recovery pattern was to replay the saved artifact instead of rerunning the whole workflow.
  - Tags: ai-agents, automation, debugging, reliability, openclaw, agent-ops
- [One Heading Level Broke the Nightly Build: Fixing Markdown Drift in Generated REM Reports](https://anyech.github.io/jingxiao-cai-blog/one-heading-level-broke-nightly-build-generated-rem-reports.html)
  - Markdown mirror: https://anyech.github.io/jingxiao-cai-blog/one-heading-level-broke-nightly-build-generated-rem-reports.html.md
  - Date: 2026-05-08
  - Summary: A generated Markdown report failed over one heading-level jump; the durable fix was testing each rendered output surface as its own artifact contract.
  - Tags: openclaw, automation, markdown, regression-testing, ai-agents, writing
- [Why Custom Skills Did Not Load in OpenClaw - A Historical Bug and Follow-Up](https://anyech.github.io/jingxiao-cai-blog/workspace-skills-bug.html)
  - Markdown mirror: https://anyech.github.io/jingxiao-cai-blog/workspace-skills-bug.html.md
  - Date: 2026-02-23; updated: 2026-05-08
  - Summary: A historical custom-skill loading bug, refreshed with a generated-Markdown regression lesson and a standalone artifact-contract follow-up.
  - Tags: openclaw, bugs, ai-agents, customization
- [Long-Running Agent Work Needs a Bridge Back, Not Just a Background Thread](https://anyech.github.io/jingxiao-cai-blog/long-running-agent-work-needs-bridge-back.html)
  - Markdown mirror: https://anyech.github.io/jingxiao-cai-blog/long-running-agent-work-needs-bridge-back.html.md
  - Date: 2026-05-07
  - Summary: Detaching long-running agent work is useful only when admission, work ownership, and final delivery all have explicit contracts.
  - Tags: ai-agents, automation, discord, reliability, workflow, openclaw
- [When SQLite Looks Empty but Isn’t: Reproducing Corrupt Task Registries Without Touching Prod](https://anyech.github.io/jingxiao-cai-blog/sqlite-empty-corrupt-task-registries-without-touching-prod.html)
  - Markdown mirror: https://anyech.github.io/jingxiao-cai-blog/sqlite-empty-corrupt-task-registries-without-touching-prod.html.md
  - Date: 2026-05-06
  - Summary: A self-hosted agent-ops debugging story: raw SQLite can still see rows while the runtime registry restore fails, so reproduce on copies before touching production.
  - Tags: sqlite, openclaw, ai-agents, incident-response, debugging, self-hosted
- [The 10-Second Session List: Why Prefiltering Before Row Build Matters in Agent Gateways](https://anyech.github.io/jingxiao-cai-blog/10-second-session-list-prefilter-row-build-agent-gateway.html)
  - Markdown mirror: https://anyech.github.io/jingxiao-cai-blog/10-second-session-list-prefilter-row-build-agent-gateway.html.md
  - Date: 2026-05-02
  - Summary: A self-hosted agent-gateway performance lesson: if a tiny session-list request builds hundreds of rich rows before filtering, limit is too late to save you.
  - Tags: openclaw, ai-agents, performance, devops, control-plane, self-hosted
- [Closing External Threads Cleanly: An Agent-Ops Pattern](https://anyech.github.io/jingxiao-cai-blog/closing-external-threads-cleanly-agent-ops.html)
  - Markdown mirror: https://anyech.github.io/jingxiao-cai-blog/closing-external-threads-cleanly-agent-ops.html.md
  - Date: 2026-05-01
  - Summary: A lightweight agent-operations pattern for closing external threads cleanly: make constraints explicit, record the decision, finish the action, and define reopen criteria.
  - Tags: ai-agents, automation, workflow, human-in-the-loop, decision-making
- [Design-Tool Integrations Need Capability Gates: Lessons from a Missing LLM Config](https://anyech.github.io/jingxiao-cai-blog/design-tool-integrations-capability-gates-llm-config.html)
  - Markdown mirror: https://anyech.github.io/jingxiao-cai-blog/design-tool-integrations-capability-gates-llm-config.html.md
  - Date: 2026-04-30
  - Summary: Why design-tool integrations need capability gates before LLM generation: validate inputs, route readiness, model config, and artifact proof early.
  - Tags: tooling, ai-agents, workflow, design-tools, llm-ops, validation
- [Treating AI Agent Updates Like Production Deployments: The Runbook Keeps Paying Off](https://anyech.github.io/jingxiao-cai-blog/ai-agent-updates-production-deployments-runbook.html)
  - Markdown mirror: https://anyech.github.io/jingxiao-cai-blog/ai-agent-updates-production-deployments-runbook.html.md
  - Date: 2026-04-30
  - Summary: Why self-hosted AI-agent updates need production-deployment discipline: preflight, backup, staged rollout, human activation, adoption scans, and verification.
  - Tags: openclaw, ai-agents, release-engineering, reliability, rollback, devops
- [Modernizing Agent Skills Without Growing a Skill Jungle](https://anyech.github.io/jingxiao-cai-blog/modernizing-agent-skills-without-growing-skill-jungle.html)
  - Markdown mirror: https://anyech.github.io/jingxiao-cai-blog/modernizing-agent-skills-without-growing-skill-jungle.html.md
  - Date: 2026-04-29
  - Summary: How I modernized agent skills with problem-first discovery, intake gates, thin wrappers, package hygiene, capability gates, and consolidation instead of skill sprawl.
  - Tags: ai-agents, openclaw, skills, workflow, maintenance, governance
- [VPS OAuth Survival Guide: Google APIs Without a Browser](https://anyech.github.io/jingxiao-cai-blog/vps-oauth-survival-guide.html)
  - Markdown mirror: https://anyech.github.io/jingxiao-cai-blog/vps-oauth-survival-guide.html.md
  - Date: 2026-02-25; updated: 2026-04-29
  - Summary: Complete tutorial on OAuth 2.0 for headless servers, now with a fail-closed readiness gate that links OAuth-backed automation checks to the broader agent-launch gate pattern.
  - Tags: tutorial, oauth, vps, devops, automation, google-cloud
- [The Nightly Build: How My Agent Runs Security Audits While I Sleep](https://anyech.github.io/jingxiao-cai-blog/nightly-build-security-audits.html)
  - Markdown mirror: https://anyech.github.io/jingxiao-cai-blog/nightly-build-security-audits.html.md
  - Date: 2026-03-02; updated: 2026-04-27
  - Summary: My AI agent runs autonomous cron jobs every night—security audits, health checks, and documentation—now updated with the exact-exec driver lesson that prevents false-negative wrapper alerts from hiding real command success.
  - Tags: ai-agents, devops, automation, openclaw, security
- [The Supply Chain Attack on AI Agents: What OpenClaw Users Need to Know](https://anyech.github.io/jingxiao-cai-blog/supply-chain-attack-ai-agents.html)
  - Markdown mirror: https://anyech.github.io/jingxiao-cai-blog/supply-chain-attack-ai-agents.html.md
  - Date: 2026-02-23; updated: 2026-04-03
  - Summary: A deep dive into malicious skills in AI agent platforms, now updated with the LiteLLM incident, the first named downstream victim report, LangChain/LangGraph vulnerabilities, exposed Ollama servers, and an approved-but-blocked lesson from Moltbook's unstable post-incident API surface.
  - Tags: security, openclaw, ai-agents, supply-chain, moltbook
- [The LiteLLM Supply Chain Attack: What OpenClaw Users Need to Know](https://anyech.github.io/jingxiao-cai-blog/litellm-supply-chain-attack-openclaw.html)
  - Markdown mirror: https://anyech.github.io/jingxiao-cai-blog/litellm-supply-chain-attack-openclaw.html.md
  - Date: 2026-03-28
  - Summary: LiteLLM 1.82.7–1.82.8 were malicious PyPI releases tied to a compromised Trivy CI/CD path. This operator-focused write-up covers the fast audit, rotation, inspection, and pinning checklist for OpenClaw users.
  - Tags: security, openclaw, ai-agents, supply-chain, python, incident-response
- [Bigger Embeddings ≠ Better Memory: Why I Chose text-embedding-3-small for OpenClaw Remote Memory](https://anyech.github.io/jingxiao-cai-blog/text-embedding-3-small-openclaw-remote-memory.html)
  - Markdown mirror: https://anyech.github.io/jingxiao-cai-blog/text-embedding-3-small-openclaw-remote-memory.html.md
  - Date: 2026-03-27; updated: 2026-03-28
  - Summary: After proving local memory search worked, I stabilized a remote memory-only lane in OpenClaw. The follow-up reinforced the same lesson: source discipline, lexical anchors, and hybrid retrieval mattered more than another round of model churn.
  - Tags: openclaw, ai-agents, memory, embeddings, debugging, devops
- [OpenClaw 2026.3.12 Regression: When logs --follow Breaks But the Gateway Stays Healthy](https://anyech.github.io/jingxiao-cai-blog/openclaw-logs-follow-regression-2026-3-12.html)
  - Markdown mirror: https://anyech.github.io/jingxiao-cai-blog/openclaw-logs-follow-regression-2026-3-12.html.md
  - Date: 2026-03-14; updated: 2026-03-19
  - Summary: After upgrading OpenClaw from 2026.3.11 to 2026.3.12, `openclaw logs --follow` failed with a misleading gateway error while the gateway stayed healthy. Updated with the 2026.3.13 resolution, local retest, and a related local-memory troubleshooting win on the same VPS.
  - Tags: openclaw, troubleshooting, regression, logs, websocket, devops
- [The Hidden Input Limit: When "202K Context" Doesn't Mean 202K](https://anyech.github.io/jingxiao-cai-blog/hidden-input-limit-llm-apis.html)
  - Markdown mirror: https://anyech.github.io/jingxiao-cai-blog/hidden-input-limit-llm-apis.html.md
  - Date: 2026-02-24; updated: 2026-03-16
  - Summary: A debugging story about hidden platform caps, now updated with Anthropic's March 2026 flat 1M Claude pricing change and why cheaper long context still doesn't eliminate API-level ceilings.
  - Tags: llm, debugging, bailian, openclaw, claude, pricing
- [The Recovery Problem: Why Your AI Agent Needs an Undo Button](https://anyech.github.io/jingxiao-cai-blog/recovery-problem-ai-agent-undo.html)
  - Markdown mirror: https://anyech.github.io/jingxiao-cai-blog/recovery-problem-ai-agent-undo.html.md
  - Date: 2026-03-07
  - Summary: I run autonomous cron jobs with no built-in undo capability. When Moltbook's community started talking about recovery primitives, I realized that unattended automation needs a stronger recovery story.
  - Tags: ai-agents, devops, recovery, openclaw, automation, safety
- [Blog Post Sanitization Checklist: What to Redact Before Publishing](https://anyech.github.io/jingxiao-cai-blog/blog-sanitization-checklist.html)
  - Markdown mirror: https://anyech.github.io/jingxiao-cai-blog/blog-sanitization-checklist.html.md
  - Date: 2026-03-03
  - Summary: I created a sanitization checklist after nearly publishing sensitive deployment details. Here's what to redact, what to keep, and validation scripts for technical bloggers.
  - Tags: writing, security, blogging, opsec, technical-writing
- [Building a Personal AI Assistant: From Gmail to Morning Memos](https://anyech.github.io/jingxiao-cai-blog/personal-ai-assistant-gmail-morning-memo.html)
  - Markdown mirror: https://anyech.github.io/jingxiao-cai-blog/personal-ai-assistant-gmail-morning-memo.html.md
  - Date: 2026-02-21; updated: 2026-03-03
  - Summary: How I built an AI assistant to automate my morning email routine.
  - Tags: ai, automation, gmail
- [Processing Background Shapes: From Radar Signals to Distributed ML Runtime Systems](https://anyech.github.io/jingxiao-cai-blog/processing-background-shapes.html)
  - Markdown mirror: https://anyech.github.io/jingxiao-cai-blog/processing-background-shapes.html.md
  - Date: 2026-02-19
  - Summary: How my PhD in radar signal processing translated into distributed ML runtime systems and backend execution work.
  - Tags: radar, signal processing, ML runtime systems, distributed systems, Oracle, HeatWave, career
- [Getting OneDrive Working with Self-Hosted AI Agents: A Survival Story](https://anyech.github.io/jingxiao-cai-blog/onedrive-multcloud-story.html)
  - Markdown mirror: https://anyech.github.io/jingxiao-cai-blog/onedrive-multcloud-story.html.md
  - Date: 2026-02-18
  - Summary: How I migrated my cloud storage using MultCloud.
  - Tags: cloud, onedrive, multcloud
- [Setting Up Google APIs for Self-Hosted AI Agents](https://anyech.github.io/jingxiao-cai-blog/google-api-setup-guide.html)
  - Markdown mirror: https://anyech.github.io/jingxiao-cai-blog/google-api-setup-guide.html.md
  - Date: 2026-02-18
  - Summary: A comprehensive guide to setting up Google APIs for your projects.
  - Tags: api, google, tutorial
- [Troubleshooting AI Agent Skills: A Debugging Story](https://anyech.github.io/jingxiao-cai-blog/troubleshooting-ai-agent-skills.html)
  - Markdown mirror: https://anyech.github.io/jingxiao-cai-blog/troubleshooting-ai-agent-skills.html.md
  - Date: 2026-02-18
  - Summary: Common issues and solutions when working with AI agent skills.
  - Tags: ai-agents, troubleshooting
- [Migrating from WordPress to GitHub Pages](https://anyech.github.io/jingxiao-cai-blog/migrating-from-wordpress.html)
  - Markdown mirror: https://anyech.github.io/jingxiao-cai-blog/migrating-from-wordpress.html.md
  - Date: 2026-02-17
  - Summary: I used to run a blog on WordPress back in 2016-2017 during my PhD years at University of Oklahoma...
  - Tags: blog, tech
- [Welcome to My Blog](https://anyech.github.io/jingxiao-cai-blog/welcome.html)
  - Markdown mirror: https://anyech.github.io/jingxiao-cai-blog/welcome.html.md
  - Date: 2026-02-17
  - Summary: After several years in industry, I'm starting to share my learnings, thoughts, and experiences...
  - Tags: intro, personal