using-leyline

<SUBAGENT-STOP> If you were dispatched as a subagent with a specific task, skip this skill and execute your task. </SUBAGENT-STOP> <EXTREMELY-IMPORTANT> If there is even a 1% chance a skill might apply to what you are doing, you MUST invoke it. If a skill applies, you do not have a choice. You must use it. </EXTREMELY-IMPORTANT> ## Instruction priority 1. **Human partner's explicit instructions** (CLAUDE.md, AGENTS.md, GEMINI.md, direct messages) - highest 2. **Leyline skills** - override default system prompt behavior where they conflict 3. **Default system prompt** - lowest If a manifest says "don't use TDD" and a skill says "always use TDD," follow the human partner. The human partner is in control. ## How to access skills - **Claude Code / Cursor:** use the `Skill` tool. The skill content is loaded and presented to you; follow it directly. Do not use `Read` on `SKILL.md` files. - **Codex / OpenCode / Copilot CLI:** use the `skill` tool (lowercase) with the skill name. - **Gemini CLI:** use `activate_skill`. Never guess a skill name from training. Only invoke skills listed in the inventory (`skills/` folders) or ones the human partner names explicitly. ## The rule ``` Before any response or action - including clarifying questions - check whether any Leyline skill applies. If one does (probability >= 1%), invoke it before narrating. If none does, name the skills you considered and why you rejected each. ``` This rule appears verbatim in `CLAUDE.md`, `AGENTS.md`, `GEMINI.md`, `README.md`, and this skill so the same instruction reaches the agent through every load path. Drift across these files is caught by `scripts/check-manifests.sh`. If the rule applies and a skill does, invoke it via the harness's skill mechanism (`Skill` in Claude Code / Cursor; `skill` in Codex / OpenCode / Copilot CLI; `activate_skill` in Gemini CLI). Even a 1% chance a skill might apply means you invoke it to check; if it turns out not to apply, you do not have to follow it. Naming the skills you considered and why is the visible-justification clause: it makes the no-skill-applies path auditable. "I checked `brainstorming` (request is a question, not a build), `systematic-debugging` (no failure surfaced), `using-git-worktrees` (no isolation needed); none applies" is acceptable. "No skill applies" alone is not. ```dot digraph skill_flow { "Message received" [shape=doublecircle]; "Might any skill apply?" [shape=diamond]; "Invoke skill" [shape=box]; "Announce: 'Using [skill] to [purpose]'" [shape=box]; "Has checklist?" [shape=diamond]; "Create per-item task entries" [shape=box]; "Follow skill exactly" [shape=box]; "Respond" [shape=doublecircle]; "Message received" -> "Might any skill apply?"; "Might any skill apply?" -> "Invoke skill" [label="yes, even 1%"]; "Might any skill apply?" -> "Respond" [label="definitely not"]; "Invoke skill" -> "Announce: 'Using [skill] to [purpose]'"; "Announce: 'Using [skill] to [purpose]'" -> "Has checklist?"; "Has checklist?" -> "Create per-item task entries" [label="yes"]; "Has checklist?" -> "Follow skill exactly" [label="no"]; "Create per-item task entries" -> "Follow skill exactly"; "Follow skill exactly" -> "Respond"; } ``` ## The pipeline - where to enter Leyline encodes a fixed-order pipeline. Map the human partner's opening message to the right entry skill. | Human partner says | Entry skill | |--------------------|-------------| | "let's build X", "I want to add Y", "we should make Z" | `brainstorming` (stage 1a) | | "debug this", "fix this bug", "something is broken" | `systematic-debugging` (6a.2) | | "review this code" | `requesting-code-review` (stage 7) | | "start implementing the plan", "execute the plan" | `subagent-driven-development` or `executing-plans` (stage 5) | | "let's plan this" (spec exists) | `writing-plans` (stage 4) | | "finish the branch", "ship this" | `finishing-a-development-branch` (stage 8) | | "resume the kept branch", "continue the feature I left" | check the kept branch's baseline note: if plan has unchecked tasks, `subagent-driven-development`; if review is incomplete, `requesting-code-review`; if ready to integrate, `finishing-a-development-branch`. Ask the human partner which state if unclear. | When the human partner says "let's build X," the pipeline is `brainstorming` -> `design-brainstorming` (when surfaces) -> `deep-discovery` -> `design-interrogation` (conditional) -> `using-git-worktrees` -> `writing-plans` -> `subagent-driven-development` or `executing-plans` -> `requesting-code-review` + `requesting-design-review` -> `finishing-a-development-branch`. Each skill names its successor explicitly. Follow the named successor, not your instinct. ### Missing-successor fallback If a skill names a successor that is not present in this version of the plugin (the harness reports the skill does not exist, or `skills/<successor>/SKILL.md` is missing), STOP. Tell the human partner the pipeline is incomplete and which successor is missing. Do not improvise the missing stage; do not skip ahead to the next named successor. The pipeline's value is determinism - missing a stage silently is worse than visibly halting. ## Red flags These thoughts mean STOP - you are rationalizing: | Thought | Reality | |---------|---------| | "This is just a simple question" | Questions are tasks. Check for skills. | | "I need more context first" | Skill check comes BEFORE clarifying questions. | | "Let me read some files first" | Skills tell you HOW to explore. Check first. | | "I'll just answer quickly" | Quick answers are how discipline slips. Check first. | | "This doesn't need a formal skill" | If a skill exists, use it. | | "I remember this skill" | Skills evolve. Read the current version. | | "This doesn't count as a task" | Any action is a task. Check first. | | "The skill is overkill" | Simple things become complex. Use it. | | "I'll check skills after this one thing" | Check BEFORE doing anything. | | "I know what to do here" | Knowing the concept is not using the skill. Invoke it. | ## Skill types - **Rigid** (for example TDD, systematic-debugging) - follow exactly. Do not adapt away the discipline. - **Flexible** (patterns) - adapt principles to context. The skill itself tells you which. ## Iron laws (always in effect) Code Discipline: - `NO PRODUCTION CODE WITHOUT A FAILING TEST FIRST` - `NO FIXES WITHOUT ROOT CAUSE INVESTIGATION FIRST` - `NO COMPLETION CLAIMS WITHOUT FRESH VERIFICATION EVIDENCE` Experience Discipline (additional, when a user-facing surface is touched): - `NO USER-FACING SURFACE WITHOUT AN APPROVED DESIGN ARTIFACT FIRST` - `NO COMPLETION CLAIMS ON A USER-FACING SURFACE WITHOUT FRESH ACCESSIBILITY EVIDENCE` > Violating the letter of the rules is violating the spirit of the rules. ## Terminology - The person you are collaborating with is the **human partner**. Not "user," not "client." - Stage 1 is **Discovery**. "Design" is reserved for UI/UX. - The 6b overlay is **Experience**. Not "Frontend" or "UI". - The artifact produced by `design-brainstorming` is a **UX artifact** (markdown). Not a "mockup" or "wireframe." ## Announce on entry The first time you invoke a Leyline skill in a session, say: > Using Leyline - checking for applicable skills before responding. Then invoke the specific skill and follow it. ## Forbidden phrases Do not say: - "Let me answer this quick question first, then check for skills." - "The skill check is overkill for this one; I'll just answer." - "Remember the skill from earlier; no need to re-read it." - "Let me gather context first, then decide whether to invoke a skill." - "I'll skip the announce; the human partner will see I'm following the skill from context." - "The successor skill is obvious; I don't need to name it explicitly." ## Successor The next skill depends on the situation. Use the table above to pick the entry skill, then follow the successor each invoked skill names.