Built-in action definitions for the Signadot platform. The directory layout is inspired by the Agent Skills spec.
Each action is a directory containing an ACTION.md file with YAML frontmatter
and a markdown body describing the action's behavior, inputs, outputs, and
executable script blocks.
The body is the action implementation as markdown with fenced code blocks. The text in the markdown may contain some directives to specify
- Runner Requirements
- Inputs and Outputs
Inline I/O declarations use \input{...} and \output{...} syntax:
\input{name}
\input{name, required}
\input{name, default="value"}
\input{name, required, schema={"type":"integer","minimum":1}}
\input{name, required, schemaRef="schema/name-schema.json"}
\output{...} supports the same options except required.
schemaRef can only refer to files relative to the directory in which
ACTION.md appears.
Script blocks are fenced code blocks that define the executable
implementation of the action. Any code block without a language tag or with sh or
bash language tags are included in the script blocks of the ACTION.
Scripts read inputs from ./context/<name> and write outputs to
./outputs/<name>. Validation blocks use the validation language tag and
are extracted separately for runner capability checks.
If the action has authoring guidance for plan authors — anti-patterns,
when-not-to-use, cross-action gotchas — put it under a top-level
## Authoring rules H2 section, between the descriptive prose / runtime
behavior and the trailing implementation script block. The convention
gives downstream consumers (e.g. the LLM compile prompt) a stable anchor
to extract just the agent-facing text without pulling in implementation
details. Actions that have no special rules can simply omit the section.
| Action | Description |
|---|---|
request-http |
Execute an HTTP request and capture the full roundtrip as structured JSON |
check |
Evaluate an Expr boolean expression against input JSON for pass/fail assertions |
eval |
Evaluate an expression against named inputs and return the result |
k6 |
Run a k6 load test and capture the summary |
playwright |
Run Playwright browser tests and capture results |