The manifest bundler and workflow-version validation each maintained
their own walk over the graph reference vocabulary (goal files, inline
goal/prompt templates, imports, child workflows, @-file attributes), so
a new reference-bearing attribute had to be added twice or the two
would drift. Add fabro_template::visit_graph_references as the single
per-graph walker — it validates that file references are template-free
and emits typed events — and rewrite both consumers on top of it,
keeping their own IO, resolution, and recursion.
Same for dockerfiles: SettingsLayer::image_layers[_mut] is now the one
definition of where images live in a settings layer, replacing the
three hand-rolled traversals in the run compiler, manifest bundler, and
workflow-version validation.
Behavior note: the manifest bundler now also follows
stack.child_dotfile (already in the shared vocabulary and treated as a
child-workflow reference by workflow-version validation); nothing in
the engine emits or reads that attribute today.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Move the WorkflowVersion wire type and its structural invariants (file
count and size limits, entrypoint presence, unique keys, path
collisions, canonical form) into fabro-types, so fabro-api replaces the
generated schema type without pulling graph parsing or the template
engine into every API consumer. The wire shape is unchanged.
fabro-workflow-version keeps the expensive graph/config/template
validation behind a ValidatedWorkflowVersion newtype and now owns
WorkflowVersionStore: put only accepts validated versions and get still
re-validates blobs read from shared storage. fabro-store goes back to
being domain-agnostic persistence.
Move the static-reference attribute vocabulary (ReferenceKind,
AttributeScope, reference_kind_for_attribute) to fabro_types::graph and
template-syntax validation to fabro-template, so fabro-graphviz no
longer depends on the template engine.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Command node `script` attributes were literal text: a `{{ inputs.x }}`
reached bash verbatim, and the only signal was a `detemplated_attribute`
warning. Scripts now substitute `{{ goal }}`, `{{ inputs.NAME }}`, and
`{{ vars.NAME }}` at run creation, alongside goals and prompts.
Scripts use `InterpString` token substitution rather than the MiniJinja
pass that renders prompts. Shell source is full of brace syntax that must
survive untouched — jq filters, awk programs, Go templates, brace
expansion — and `InterpString` claims only the narrow token forms,
leaving everything else literal.
`env` and `secrets` are deliberately not wired and now fail loudly
instead of passing through as text. A script reads the environment with
`$NAME`, which needs no interpolation, and a resolved secret would be
baked into the `CommandStarted` event that records the script verbatim.
The error points at `[environments.<slug>.env]` for the secret case.
`ResolveCtx` gains opt-in `with_inputs` and `with_goal`. Namespace
availability stays scope-determined per call site, so every existing
config-layer context leaves both unwired and keeps its current behavior.
`goal` names a single value rather than a namespace of them, so it has
no dotted form: only the exact body `goal` produces a token and
`{{ goal.title }}` stays literal.
Values substitute verbatim without shell quoting, matching
`[[run.prepare.steps]].script` where the snippet is the author's to
quote. Substituted text is never rescanned.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
