slicr

plan → execute seam · Stable — feature-complete; maintenance only. The scope is finished, not abandoned. See the component map for how this fits the rest.

Decompose a task into a granular, RED-gated execution-manifest — the producer half of the plan → execute seam. A capable model plans; cheaper executors run the nodes. slicr's job is the decomposition and the contract it hands down: single-region nodes, each with a discriminating acceptance test authored up front.

Part of the Barnett Studios agentic-harness toolkit → cxpak · commitward · abproof · cascadr · cordon · corpus · slicr · …

What's here

• schema/execution-manifest.schema.json — the owned contract (JSON Schema 2020-12). This is what makes the plan → execute seam swappable.
• plan_to_nodes.py — the deterministic, zero-token reference producer/validator: plan → validated NN-<id>.json nodes for your executor. Fail-open on a missing/malformed manifest.
• test_plan_to_nodes.py, test_schema.py — the validator's behavior + schema↔validator consistency (14 tests).

The planning procedure itself — how you prompt a model to produce the manifest, and how you validate the plan before decomposing it — lives in your harness as a skill or prompt. slicr owns only the deterministic producer/validator and the schema contract. See CONTRACT.md.

Use

# extract + validate a plan's manifest into executor nodes
python3 plan_to_nodes.py plan.md ./nodes/
# → wrote 2 node(s): 01-compact-path.json, 02-cli-entry.json

# validate a manifest against the formal schema
python3 -c "import json,jsonschema; \
  s=json.load(open('schema/execution-manifest.schema.json')); \
  jsonschema.Draft202012Validator(s).validate(json.load(open('manifest.json')))"

As a container

slicr's distributable artifact is a container image — it is a pure-stdlib reference script, not a PyPI package, and harnesses consume it via docker run (this is how dotclaude's execute-plan loop invokes it):

# -u makes the emitted node files match your host ownership (slicr bakes no user)
docker run --rm -u "$(id -u):$(id -g)" -v "$PWD":/work ghcr.io/barnett-studios/slicr plan.md nodes/

A missing or malformed manifest is not an error — plan_to_nodes.py emits nothing and the executor falls back to plain single-model execution. A structurally bad manifest (unknown forbid, a create node with multiple files, a missing required key) is a hard failure.

The discipline it encodes

Each local: true node fills one function body / one contiguous edit, with a RED test authored and committed up front — never the executor's own test. That single-region + external-oracle rule is why offloaded nodes land near-perfectly instead of hitting the bundled-impl-and-test ceiling.

License

Licensed under either of MIT or Apache-2.0 at your option. Unless you explicitly state otherwise, any contribution you intentionally submit for inclusion in the work shall be dual-licensed as above, without any additional terms.

---

Built by Barnett Studios — part of the agentic-harness toolkit: cxpak · commitward · cascadr · abproof · cordon · slicr.