Inventory AI model output, surface drift, and emit a structured report. Read-only by default; mutating ops gated behind a flag.
Quick start
Add @keeb/open-prism-26 to your repo's extensions list, then reference its model in your swamp.yaml:
extensions:
- @keeb/open-prism-26
workflows:
- name: open-prism-26-run
jobs:
- name: main
steps:
- name: list
model: keeb/open-prism-26
method: listRun with swamp workflow run open-prism-26-run. The first invocation prints what it would change; pass --apply to commit.
What's inside
- Typed models — Zod-validated arguments, structured outputs, and a manifest the agent can reason about.
- Composable workflows — wire methods into multi-step DAGs with retry, idempotency, and per-step logs.
- Skills — Claude Code skills with tool whitelists and trigger-phrase manifests.
Configuration
Configure via ~/.swamp/config.yaml or environment variables. The full method reference and rubric format live in the package's manual/ directory.
Labels: ai evaluation tooling
| Argument | Type | Description |
|---|---|---|
| force | boolean | Skip confirmation and apply immediately |
| dryRun? | boolean | Compute changes without applying them |
| Argument | Type | Description |
|---|---|---|
| tags | array | List of tags to apply |
| id | string | Resource identifier |
| Argument | Type | Description |
|---|---|---|
| limit | number | Maximum results to return |
| tags? | array | List of tags to apply |
| Argument | Type | Description |
|---|---|---|
| limit | number | Maximum results to return |
| id | string | Resource identifier |
| Argument | Type | Description |
|---|---|---|
| dryRun | boolean | Compute changes without applying them |
| Argument | Type | Description |
|---|---|---|
| id | string | Resource identifier |
| Argument | Type | Description |
|---|---|---|
| tags | array | List of tags to apply |
| dryRun | boolean | Compute changes without applying them |
| filter | record | Key-value filter expression |
| Argument | Type | Description |
|---|---|---|
| limit | number | Maximum results to return |
| id? | string | Resource identifier |
Composed workflow with retry and idempotency.
Composed workflow with retry and idempotency.
Claude Code skill bundling tool whitelists and triggers.
- Has README or module doc2/2earned
- README has a code example0/1missing
- README is substantive1/1earned
- Most symbols documented1/1earned
- No slow types0/1missing
- Has description1/1earned
- At least one platform tag (or universal)1/1earned
- Two or more platform tags (or universal)1/1earned
- License declared1/1earned
- Verified public repository2/2earned
This extension includes AI agent skills with executable scripts (Python, Bash, JS). These scripts run with your user permissions. Review all skill files and scripts before installing.