All skills
simota avatar

/loom

@8e1f365
by shingo imotasimota/agent-skills85 stars
15

Designing a repository's operating layer — which project-local skills, recipes, and orchestration workflows it needs, as one system. Not for a single skill body (Sigil) or runtime routing (Nexus).

Use this Skill: https://skilld.dev/gh/simota/agent-skills/loom

This session only. Nothing lands on disk.

referencerecipe-design.md

≈1.2k tokens on demand. Your agent reads this file only when SKILL.md points to it.

Recipe Design

Purpose: How to design a repo-tailored task playbook as a subcommand-driven recipe. Read when: You are in the recipe mode or designing the recipe set inside a blueprint.

Contents

  • What a recipe is (and is not)
  • Recipe anatomy
  • Signal keywords
  • Worked examples
  • Anti-patterns

What a recipe is (and is not)

A recipe is a named, repeatable playbook for one recurring multi-step task in the repo. It chains existing skills and ecosystem agents in a fixed order with decision points. It is the project-scoped analogue of the Recipe tables that ecosystem skills (Architect, Sigil) carry internally.

A recipe IS A recipe is NOT
A fixed flow for a repeatable task A single methodology (→ that is a skill)
A chain of skills/agents with a subcommand Runtime per-request routing (→ that is Nexus)
Design-time, persistent, repo-local An autonomous loop (→ that is Orbit)
Deterministic enough to name and document A one-off ad-hoc sequence

A recipe earns existence only when SURVEY found the task happening repeatedly. If it happens once, do not encode it.


Recipe anatomy

Every designed recipe specifies these fields (handed to Nexus to register, and to Sigil if any step needs a new skill):

recipe:
  subcommand: ship-feature          # first-token trigger
  task: "Take a described feature from scaffold to an opened PR"
  signal_keywords: [ship, "build the feature", "implement and open PR"]
  preconditions: [clean working tree, feature spec exists]
  chain:                            # ordered steps, each with an owner
    - step: scaffold structure       owner: scaffold       gate: none
    - step: implement logic          owner: builder        gate: tests-green
    - step: add edge-case tests      owner: radar          gate: coverage>=target
    - step: prepare PR               owner: guardian       gate: user-confirm
  decision_points:
    - at: implement logic
      if: "ambiguous requirement"
      then: ask user, do not guess
  output: "Opened PR with passing tests"
  rollback: "Abandon branch; no merge performed"

The discriminator is shape, not step count: a straight chain is a recipe regardless of length; any genuine branching, fan-out, or convergence makes it a workflow — design it with workflow-design.md and a formal topology instead, even below 5 steps. Keep recipe chains short for readability, but length alone never forces promotion.


Signal keywords

Recipes activate by subcommand (first token) or by natural-language signal. For each recipe, define both so Nexus can route. Make signals concrete and non-overlapping with other recipes in the suite — two recipes that fire on the same phrase is a routing fault caught in VERIFY.

| Subcommand   | Signal keywords                          |
|--------------|------------------------------------------|
| ship-feature | "ship", "build and open PR", "implement X"|
| cut-release  | "release", "tag a version", "ship to prod"|
| onboard      | "set up dev env", "onboard", "first run" |

Worked examples

ship-feature (chain over skills + agents)

SURVEY found: features routinely go scaffold → implement → test → PR, done inconsistently. Mechanism: recipe (repeatable multi-step, no heavy coordination). Chain: scaffold → builder → radar → guardian, gate on tests-green before PR. Owner: Nexus registers it; no new skill needed.

cut-release (chain with a confirmation gate)

SURVEY found: monthly release with changelog + notes + tag. If it needs ≤ 5 linear steps → recipe; if it fans out (collect across packages, parallel verify, then converge) → promote to workflow. Decision point: ask user before tagging. Owner: Nexus; defer changelog/notes to Launch if that ecosystem agent owns it (route, don't duplicate).

add-migration (recipe that needs a new project skill)

SURVEY found: schema changes need a migration + model update + test, repo-specific. Mechanism: recipe whose first step calls a new project skill new-prisma-migration. DELEGATE: Sigil authors the skill body; Loom registers the recipe with Nexus referencing it.


Anti-patterns

  • Recipe that is really one skill. A single methodology with no chaining belongs in a skill (Sigil), not a recipe.
  • Recipe that is really a workflow. Heavy multi-agent coordination with branching needs a formal topology — promote it.
  • Overlapping signals. Two recipes firing on the same phrase. Disambiguate or merge.
  • Encoding a one-off. No observed repetition in SURVEY → do not create the recipe.
  • Embedding step logic Loom should delegate. Loom names the step and its owner; the owning skill/agent holds the how.

Source: SKILL.md on GitHub

No alerts5mo4 checks · Risk SAFE
  • Gen Agent Trust Hub5mo

    The 'Loom' skill is a tool designed for synchronizing codebases with Figma design systems by generating guidelines and prompt strategies. No security risks, malicious behaviors, or unauthorized data access patterns were detected during the analysis.

  • Socket5mo

    No alerts

  • Snyk5mo

    Risk: LOW · No issues

  • ZeroLeaks5mo

    Score: 93/100 · 2 sections analyzed

Signed by skilld at 8e1f365. This ties the file your Agent reads to that commit on GitHub. It does not review the instructions.

Last checked against GitHub 3 days ago.

Activeupdated last month

README badge

README badge for simota/agent-skills/loom