Skip to content

soulike/knowledge-base

v2026.11.5-1

A personal knowledge base with reusable Agent workflows.

contribute-to-knowledge-base

Publish a sanitized contribution proposal to the canonical knowledge base from outside its source checkout. Use when the user wants to submit a proposed change to its Knowledge, Skills, references, or Agent guidance through the installed plugin.

design-image-generation-prompts

Design a new image-generation prompt for a people-focused still image. Use when the user wants to create a drawing or image-generation prompt and no existing prompt or generated result is being evaluated.

design-image-to-video-prompts

Design a new image-to-video prompt for a people-focused continuous shot. Use when the user wants to generate a video from an image and no existing prompt or generated video is being evaluated.

design-operation-failure-recovery

Use when designing, implementing, substantively changing, or reviewing an operation or workflow whose outcome an external consumer can observe, including background work. Establish non-success outcomes and recovery even when the request only describes success.

evaluate-library-adoption

Evaluate established solutions before creating or extending reusable, domain-independent functionality. Use when planning or implementing a reusable technical capability that could be provided by an established solution.

handle-code-review-feedback

Use when deciding whether and how to address received code review feedback on work under change, with or without a pull request.

improve-dev-documentation

Audit and improve maintained software-development documentation, including standalone documents, source comments and docstrings, examples, prompts, Agent instructions, and Skills. Use when the task starts from an open-ended concern about one or more existing artifacts or an agreed quality improvement.

improve-image-generation-prompts

Improve a people-focused still-image generation attempt. Use when the user wants to review, diagnose, or improve an existing image-generation prompt, an unsatisfactory generated image, or both.

improve-image-to-video-prompts

Improve an image-to-video generation attempt for a people-focused continuous shot. Use when the user wants to review, diagnose, or improve an existing image-to-video prompt, generated video, or both.

load-collaboration-guidance

Use at the beginning of every task, alongside any task-specific Skills, to load independent collaboration principles before planning or acting. Also use when resuming without the guidance in context.

load-knowledge-catalog

Load the catalog of curated Knowledge. Use at the beginning of every task before planning or acting, and whenever the Agent needs to discover whether relevant Knowledge is available.

probe-web-usability-with-agents

Use when asked to find usability risks by having Agents attempt tasks in an interactive web prototype or existing site, or when designing or changing a critical interactive web task flow after an end-to-end path becomes operable.

propose-knowledge-base-contribution

Use at a substantive task's stable stopping point to assess whether its evidence supports a reusable knowledge-base contribution, or when the user asks for that assessment, requests a proposal, or accepts an offer to prepare one.

remediate-npm-security

Remediate npm dependency-security findings through supported upgrades or removal of proven-unused dependency chains. Use when the user asks to fix a security finding in an npm dependency graph.

resolve-invoked-codex-skill

Use only in Codex whenever a specific Skill is invoked or required by exact name, its instructions are unavailable in the current context, and the Agent would otherwise state that the Skill is absent or unavailable.

retry-via-local-proxy

Retry failed proxy-capable network requests through a detected local HTTP or SOCKS5 proxy. Use when a direct proxy-capable request fails with a connection, DNS, timeout, TLS, or HTTP 403 error; for an HTTP(S) Git clone, use only after a timeout.

review-and-improve-tests

Review and improve automated tests and their execution, diagnosing each problem at its owner. Use when reading or running automated tests to evaluate, diagnose, or improve them or their execution.

review-security

Review code, configuration, dependencies, and security fixes for vulnerabilities. Use when reviewing security-sensitive behavior or a security finding, investigating a suspected vulnerability or trust-boundary failure, or validating a security remediation.

wait-for-user-answer

Use only in Codex whenever you are about to ask the user a question or resume with an unanswered question. Keep the question pending without choosing an answer after a timeout.

watch-pull-request

Use when the user asks to watch or resume watching a pull request through ongoing review and CI activity.

write-dev-documentation

Write or update maintained software-development documentation, including standalone documents, source comments and docstrings, examples, prompts, Agent instructions, and Skills. Use when the task starts from a known need for one of these artifacts or an engineering change whose documentation impact must be assessed and addressed.

write-effective-tests

Decide whether behavior needs automated protection and write effective tests when it does. Use before proposing, adding, or expanding automated tests.