Thought-Retriever
WarnAudited by ClawScan on May 10, 2026.
Overview
The skill’s memory purpose is coherent, but it embeds an API key, sends conversation content to an external LLM, and persists conversation-derived memories with limited user controls.
Install only if you are comfortable with conversation content being sent to the DashScope/百炼-compatible API and with extracted memories being saved for future reuse. Before enabling any post-turn hook, replace the hard-coded API key with your own scoped credential, confirm the provider’s data policy, review or disable automatic memory writes, and verify the local ontology dependency and workspace path.
Findings (5)
Artifact-based informational review of SKILL.md, metadata, install specs, static scan signals, and capability signals. ClawScan does not execute the skill or run runtime probes.
Requests may be made under someone else’s exposed provider account, and the embedded key can be copied, abused, or revoked without the user’s control.
The source embeds and uses a bearer API key even though the registry declares no primary credential or environment-variable requirement, making the provider account boundary and rotation model unclear.
SF_KEY = "sk-b841f4b7c91d40ddb12502462708f361" ... headers={"Authorization": f"Bearer {SF_KEY}"Remove the hard-coded key; require a user-supplied, scoped API key via an environment variable or secure credential store, and document the provider account used.
Conversation content, including potentially sensitive information, can leave the local workspace and be processed by the external provider, especially if configured as an automatic post-turn hook.
The skill sends the user query and generated answer to an external LLM endpoint and explicitly disables environment proxy settings, which affects data-boundary and network-control expectations.
SF_API = "https://dashscope.aliyuncs.com/compatible-mode/v1/chat/completions" ... 用户问题：{user_query}\n回答内容：{generated_answer} ... session.trust_env = FalseMake external transmission explicit at install and runtime, allow opt-in or per-session disablement, redact sensitive content where possible, document provider retention/privacy terms, and avoid bypassing user-configured proxy controls unless clearly requested.
Private or incorrect information from a conversation may be converted into long-lived memory and reused in future tasks.
The documented workflow persistently stores conversation-derived thoughts and original-question snippets after conversations, but does not define retention, deletion, exclusions, or user review controls.
在 OpenClaw 配置中注册为 post-turn 钩子，每次对话结束自动触发。 ... graph.jsonl ← Thought 实体存储 ... 每个 Thought 都有 `query` 字段，记录生成它时的原始问题
Require explicit opt-in for automatic memory writes, provide review-before-save and delete/retention controls, avoid storing sensitive query text by default, and mark memories with provenance and confidence.
A mistaken or adversarially influenced memory could become more trusted over time and affect future agent behavior.
Persisted, automatically confidence-adjusted thoughts are described as inputs to later retrieval and self-evolution workflows, so a bad or poisoned thought can propagate beyond the original conversation.
置信度会在后续相关查询被触发时自动提升 ... Evolver | 读取 Thoughts，作为自我进化的分析素材
Add validation before increasing confidence, keep an audit trail, support rollback, and require user approval before memories are used for self-evolution or broad behavior changes.
The skill may fail or interact with an unexpected workspace on other systems, and its safety also depends on the locally installed ontology module.
The runtime depends on local ontology code at a hard-coded workspace path that is not included in this skill’s manifest, so the executed helper code and target workspace must be trusted separately.
WORKSPACE = Path("C:/Users/89627/.openclaw/workspace").resolve() ... sys.path.insert(0, str(WORKSPACE / "skills" / "ontology" / "scripts")) ... from ontology import load_graphUse the current user’s configured OpenClaw workspace path, declare the ontology dependency clearly, and verify the local dependency before enabling the skill.