Claude
Skill
writing-great-skills
Write, revise or review an agent skill (SKILL.md package).
Virus-scanned
Reviewed automatically before listing.
Download
sgaabdu4-building-flutter-apps-.agents_skills_writing-great-skills-c396097.zip · 3 KB
Install
skills CLI
npx skills add https://github.com/sgaabdu4/building-flutter-apps/tree/main/.agents/skills/writing-great-skills
Claude Code
claude plugin marketplace add https://llmmart.ai/marketplace.json && claude plugin install sgaabdu4-building-flutter-apps@llmmart
Git
git clone https://github.com/sgaabdu4/building-flutter-apps.git
The skills CLI installs just this skill, for any of its supported agents. Claude Code installs the whole sgaabdu4/building-flutter-apps collection as a plugin from our marketplace. Git is the plain clone.
Skill manifest
Writing Great Skills
- Scope = user request + settled constraints + existing authorization; skill adds no authority.
- Agent prose = minimum tokens + unambiguous meaning. Use
=,→,+, terse bullets/tables where clear; delete filler + repetition. Preserve conditions, actions, exceptions and proof; cryptic abbreviations ≠ useful compression.
Routes
Load matching routes only; Core always applies.
flowchart LR C[Core] -->|Package / metadata / invocation| P[references/package.md] C -->|Observed skill failure| R[references/repair.md] click P "references/package.md" click R "references/repair.md"
Core
| Check | Required outcome |
|---|---|
| Need | Removal loses reusable, non-default behavior; omit generic agent advice. |
| Trigger | Purpose + distinct use; relevant prompts match, unrelated prompts do not. Preserve implicit selection unless user requests explicit-only. |
| Disclosure | SKILL.md = routes + essential shared rules. Conditional workflows/examples/detail → references. Each route = load condition + direct link; load only applicable routes. Keep short universal rules inline. |
| Diagram | Flow, decision, state or dependency → Mermaid by default. Terse labels; one representation, no prose duplicate. Non-relational rule → bullet/table. |
| Split | Same actions + proof → keep together. Split for independent use or observed failure unresolved by clearer completion. |
| Resource | Script = repeated fragile logic; reference = conditional knowledge; asset = copied output. Each needs a current consumer; no mandatory file count. |
| Ownership | Rule + definition + caveat = one owner. Inspect package + relevant siblings/project instructions; consolidate duplicates + remove stale rules. |
| Steering | State target action + observable completion; likely mistake → replacement action. |
| Validation | Available native validator + changed links/tools/skills exist + metadata agrees. Substantial/risky change → representative trigger/non-trigger prompts + actual outcome proof. Run changed scripts; wording tests ≠ behavior proof. |
Files (building-flutter-apps)
-
agents
-
openai.yaml 211 B
interface: display_name: "Writing Great Skills" short_description: "Write terse, routed agent skills" default_prompt: "Use $writing-great-skills to author or review a terse, correctly routed agent skill."
-
-
references
-
package.md 2 KB
# Package review Scope = frontmatter + `SKILL.md` + linked references + scripts/assets + `agents/openai.yaml` when present. Apply this skill to its own edits too. | Surface | Check | | --- | --- | | Frontmatter | `name` + concise `description`; purpose + distinct trigger first. | | References | Clear load condition + valid direct link; conditional detail stays out of entrypoint. Load only applicable routes. | | Scripts/assets | Current consumer + concrete need; scripts execute successfully, assets serve actual output. | | `agents/openai.yaml` | Optional UI/invocation/tool metadata; preserve useful existing fields. Add only for a current UI/configuration need. | | UI fields | `display_name`, concise `short_description`, `default_prompt` explicitly invoking `$skill-name`; keep consistent with actual behavior. | | Invocation | Use host-native `$`/`/` invocation; explicit-only only when user requests. Preserve existing policy unless change authorized. | | Client policy | Codex: `agents/openai.yaml` → `policy.allow_implicit_invocation`. Claude: frontmatter → `disable-model-invocation`. `user-invocable` controls user access/menu visibility, not automatic selection. Verify target-host semantics. | | Dependencies | Referenced skills/tools/files exist in target environment; no unavailable or machine-specific prerequisites disguised as portable guidance. | | Migration | Review the whole package; classify keep/combine/discard by behavior, not file count. Preserve intent; remove stale commands + obsolete integration machinery. | Validation = native validator + metadata/link checks + relevant script/behavior proof. Supported client field rejected → narrow schema/type fix at validator owner; retain unknown-field rejection. Substantial trigger change → matching + nearby nonmatching requests. Report untested behavior; structural success ≠ effective invocation. Policy sources: [Codex](https://developers.openai.com/codex/skills/) · [Claude](https://code.claude.com/docs/en/skills). -
repair.md 720 B
# Repair observed skill failures Input = request + observed failure + intended outcome. Fix the smallest owner. ```mermaid flowchart TD F{Failure} F -->|Wrong task| T[Fix trigger or invocation] F -->|Missed detail| M[Sharpen route; shared rule inline] F -->|Excess context| E[Route conditional detail] F -->|Stops early| S[Clarify completion; split only if still failing] F -->|Conflict| O[One owner; delete duplicates] F -->|Cryptic| C[Restore missing meaning] T & M & E & S & O & C --> R[Rerun original + nearby valid request] R --> V{Original fixed + nearby valid?} V -->|Yes| D[Done] V -->|No| F V -->|Unknown| G[Report missing evidence] ``` New universal rule requires demonstrated need.
-
-
SKILL.md 2.2 KB
--- name: writing-great-skills description: Write, revise or review an agent skill (SKILL.md package). --- # Writing Great Skills - Scope = user request + settled constraints + existing authorization; skill adds no authority. - Agent prose = minimum tokens + unambiguous meaning. Use `=`, `→`, `+`, terse bullets/tables where clear; delete filler + repetition. Preserve conditions, actions, exceptions and proof; cryptic abbreviations ≠ useful compression. ## Routes Load matching routes only; Core always applies. ```mermaid flowchart LR C[Core] -->|Package / metadata / invocation| P[references/package.md] C -->|Observed skill failure| R[references/repair.md] click P "references/package.md" click R "references/repair.md" ``` ## Core | Check | Required outcome | | --- | --- | | Need | Removal loses reusable, non-default behavior; omit generic agent advice. | | Trigger | Purpose + distinct use; relevant prompts match, unrelated prompts do not. Preserve implicit selection unless user requests explicit-only. | | Disclosure | `SKILL.md` = routes + essential shared rules. Conditional workflows/examples/detail → references. Each route = load condition + direct link; load only applicable routes. Keep short universal rules inline. | | Diagram | Flow, decision, state or dependency → Mermaid by default. Terse labels; one representation, no prose duplicate. Non-relational rule → bullet/table. | | Split | Same actions + proof → keep together. Split for independent use or observed failure unresolved by clearer completion. | | Resource | Script = repeated fragile logic; reference = conditional knowledge; asset = copied output. Each needs a current consumer; no mandatory file count. | | Ownership | Rule + definition + caveat = one owner. Inspect package + relevant siblings/project instructions; consolidate duplicates + remove stale rules. | | Steering | State target action + observable completion; likely mistake → replacement action. | | Validation | Available native validator + changed links/tools/skills exist + metadata agrees. Substantial/risky change → representative trigger/non-trigger prompts + actual outcome proof. Run changed scripts; wording tests ≠ behavior proof. |
Comments (0)
Sign in to join the conversation.
Reviews (0)
No reviews yet.
No comments yet.