tesserakit-recipes 0.4.0

Recipes job pack for Tessera: compile multi-step workflow recipes into validated, ordered assets.

tessera-recipes

Compile multi-step workflow recipes into validated, dependency-ordered assets.

tessera-recipes reads a directory of recipe files (frontmatter + body), normalizes them into a canonical Recipe schema, validates the step dependency graph (cycle detection, dangling references, reachability), computes a topological execution order, and emits a catalog plus machine and human execution plans.

Input shape

Each recipe is either a single file <name>.recipe.md or a folder <name>/RECIPE.md. The file is YAML frontmatter followed by a markdown body.

---
name: release-notes
description: Build release notes from commits, then summarize and publish.
version: 1.0.0
tags: [git, release]
inputs:
  - name: since_tag
    type: string
    required: true
outputs:
  - name: published_url
steps:
  - id: collect
    uses: changelog-generator
    inputs:
      range: "${inputs.since_tag}..HEAD"
    produces: raw_changelog
  - id: summarize
    needs: [collect]
    inputs:
      text: "${steps.collect.output}"
    produces: summary
  - id: publish
    inputs:
      body: "${steps.summarize.output}"
    produces: published_url
---
Human-readable narrative of the workflow goes here.

Step dependencies are the union of explicit needs and edges inferred from ${steps.X} references in inputs, so authors can rely on either form.

Compile a recipe pack

tessera recipes compile --input examples/recipes/ --output ./out/recipe_pack

Artifacts written:

index.jsonl              canonical Recipe rows
plans.jsonl              machine execution plan per recipe (topo order, edges, cycle)
index.md                 human-readable catalog
validation_report.md     graph + frontmatter findings
coverage_report.md       step counts, acyclic ratio, tag distribution
execution_plans.md       per-recipe topological order + dependency edges

Validation rules

Frontmatter / structure:

• missing_name, non_canonical_name, invalid_version, missing_description, short_description, no_steps

Steps and graph:

• missing_step_id, duplicate_step_id
• self_dependency — a step depends on or references itself
• dangling_needs — needs names a step that does not exist
• dangling_step_reference — ${steps.X} points to a missing step
• dangling_input_reference — ${inputs.X} references an undeclared input
• cyclic_dependency — the step graph contains a cycle (reports the cycle path)
• unproduced_output — a declared output is not produced by any step
• unreachable_step — a step produces nothing and nothing depends on it

Cross-recipe:

• name_collision — two recipes share a name