跳到正文
cccskills
免费在 GitHub 公开

keep-why-autostart-examples

Use when skill or knowledge activation is unreliable, or setup reaches activation reliability. Not for remote, credential, publish, deploy, or irreversible changes.

中文简介正在准备中。当前显示原文说明。

查看安装方法

包含的文件(2)

  • SKILL.md4.4 KB
  • agents/openai.yaml175 B

SKILL.md(原文)

安装前可以查看交给 AI 代理的指令内容。

Keep why autostart examples

Contract

FieldBound contract
TriggerUser asks to make skill/knowledge activation more reliable in their agent tool, or first-run wizard reaches the activation-reliability question.
AuthorityReversible local: writes only named local hook configuration artifacts; rollback is undo. No remote mutation.
Side effectProject-scoped activation hook configured. Reference implementation: Claude Code SessionStart hook that injects a reminder only when a keep-the-why:config marker is present, measured 0/10 to 10/10 on affected eval cases.
DoneActivation measured improved via before/after eval on affected cases; hook scoped by config-marker presence (not unconditional); honest limits stated for tools without a hook mechanism.

Inputs

  1. Agent platform (required): the agent tool whose activation mechanism is being improved (e.g., Claude Code, Cursor, Windsurf). Must be supplied.
  2. Target skills or knowledge items (required): the set of skills or knowledge entries whose activation reliability is the goal. Must be supplied.
  3. Existing eval cases (optional): prior activation eval results to establish the before baseline. If absent, run the eval set once before configuring the hook.

Procedure

  1. Identify the agent platform's native hook or session-start mechanism. If the platform exposes no hook mechanism, state this limitation explicitly and stop: do not invent a workaround. Done when: the platform's hook mechanism is identified or the limitation is declared and the skill stops.
  2. Inventory the target skills/knowledge items and confirm each has at least one eval case that tests whether it activates when expected. Done when: every target item has at least one eval case.
  3. Run the eval set once without any hook to record the before-metric (expected: near 0/n on affected cases where activation was unreliable). Done when: the before-metric is recorded.
  4. Design a config marker (e.g., keep-the-why:config) that scopes the hook to fire only when present in the project configuration. The marker must be a project-local artifact, not a global or user-wide setting. Done when: one config marker is designed and is project-local.
  5. Implement the activation hook: at session start, when the config marker is present, inject a concise reminder that lists the target skills/knowledge items and their activation triggers. The hook must not fire when the marker is absent. Done when: the hook is implemented and fires only when the marker is present.
  6. Run the eval set again with the hook active to record the after-metric. Done when: the after-metric is recorded.
  7. Compare before and after metrics. Report the exact numbers. Done when: before and after numbers are reported.
  8. If the platform supports hooks but the hook mechanism differs from the reference implementation (Claude Code SessionStart), adapt the hook to the platform's native form while preserving the config-marker scoping pattern. Done when: the adapted hook preserves the config-marker scoping pattern.

Failure and recovery

Failure classDetectionResponse
Platform has no hook mechanismNo session-start, hook, or lifecycle event API foundDocument the limitation. Report the platform name and what was checked. Stop: do not claim the hook was configured.
Hook fires unconditionallyHook activates on projects without the config markerFix the scoping logic before measuring. An unconditional hook is a scope violation.
Eval shows no improvementAfter-metric equals before-metricReport the actual numbers. Do not claim success. Adjust the hook content or scoping and re-measure, or report non-convergence after two attempts.

Partial results: if the hook improves some but not all affected cases, report per-case metrics. Do not average away failures.

Rollback: remove the config marker and the hook configuration file. No other artifacts are modified.

Output

A report containing before-metric, after-metric, hook configuration (config marker artifact and hook implementation or reference to the platform-native hook), and platform-limitation statements where applicable.

评价

还没有评价。欢迎分享你的使用感受。

评论

无需登录即可发表。欢迎分享使用感受、提问,或说明在哪些环境下无法运行。

还没有评论。来写第一条吧。

提交后会立即公开。请勿填写邮箱、电话等个人信息。URL 最多 2 个。

0 / 1000 字

此表单受 reCAPTCHA 保护,适用 Google 的隐私权政策和服务条款。

同一仓库的技能

Use when explaining System V AMD64, ARM AAPCS, RISC-V psABI, stack frames, variadic calls, or FFI register rules. Not for the Rust FFI binding layer: use rust-ffi.

中文简介正在准备中。当前显示原文说明。

OutlineDriven/odin-claude-plugin372026年9月29日 更新

Use when configuring ADC sampling time, DMA-driven ADC, calibration, or DAC channel setup on bare-metal MCUs. Not for the DMA stream itself: use dma-baremetal.

中文简介正在准备中。当前显示原文说明。

OutlineDriven/odin-claude-plugin372026年9月29日 更新

af-xdp

免费

Use when creating AF_XDP sockets, configuring UMEM and XSK rings, writing an XDP redirect program, or choosing copy versus zero-copy mode. Not for full kernel bypass: use dpdk.

中文简介正在准备中。当前显示原文说明。

OutlineDriven/odin-claude-plugin372026年9月29日 更新

Use when a completed session needs an agent-environment retrospective. Not for an engineering retrospective from telemetry: use engineering-retrospective.

中文简介正在准备中。当前显示原文说明。

OutlineDriven/odin-claude-plugin372026年9月29日 更新

Use when a redacted, trimmed agent transcript must be appended to a GitHub PR or issue body, with human approval and preview. Not for automated or model-initiated insertion.

中文简介正在准备中。当前显示原文说明。

OutlineDriven/odin-claude-plugin372026年9月29日 更新

agents-md

免费

Use when a repo needs agent setup, AGENTS.md added or made lean, CLAUDE.md audited, or agent instructions scored or pruned. Not for remote, credential, publish, deploy, or irreversible changes.

中文简介正在准备中。当前显示原文说明。

OutlineDriven/odin-claude-plugin372026年9月29日 更新

查看 OutlineDriven 的全部技能

报告此技能的问题