@interf/compiler 0.3.4 → 0.4.0

package/skills/benchmark/SKILL.md

@@ -1,122 +0,0 @@
----
-{
-  "name": "benchmark",
-  "description": "Compare compiled workspaces with Interf Compiler's file-based benchmark specs. Use when asked to benchmark, compare workflows, compare compiled workspaces, or run checks on already-compiled Interf Compiler targets."
-}
----
-
-# Benchmark Compiled Workspaces
-
-Use this skill when the job is to compare compiled outputs built by Interf Compiler, not to build them from scratch.
-
-Prefer the CLI:
-
-```bash
-interf test
-```
-
-This skill is for:
-- comparing compiled workspaces built from the same source folder
-- creating or refining simple benchmark specs under `interf/benchmarks/`
-- summarizing which workflow or compiled workspace performed better on deterministic checks
-
-This skill is **not** for:
-- compiling a workspace
-- inventing a second benchmark system outside Interf Compiler's filesystem contract
-
-## Source of truth
-
-- benchmark specs in `interf/benchmarks/`
-- saved benchmark runs in `interf/benchmarks/runs/`
-- current target eligibility from `interf status`
-- public command: `interf test`
-- technical alias: `interf benchmark`
-
-When this skill is used inside a generated workspace, do not try to open SDK repo source files outside the current workspace. The benchmark spec files and saved run artifacts are the source of truth.
-
-## Benchmark model
-
-Interf Compiler benchmark specs are file-first and deterministic by default.
-
-They live under `interf/benchmarks/`.
-
-Saved benchmark runs live under `interf/benchmarks/runs/workspace/`.
-
-Minimal example:
-
-```json
-{
-  "type": "workspace",
-  "name": "Weekly Brief Quality",
-  "description": "Compare weekly briefing workspaces.",
-  "cases": [
-    {
-      "id": "current-state",
-      "question": "Does this workspace clearly say what changed and what matters next?",
-      "file": "home.md",
-      "expect": {
-        "must_include": ["launch", "people"],
-        "must_not_include": ["TODO"],
-        "min_words": 5
-      }
-    }
-  ]
-}
-```
-
-Interpretation:
-- `question` = plain-English assertion the user cares about
-- `file` = output artifact to inspect
-- `expect` = deterministic checks
-
-Do not replace this with opaque judge prompts by default.
-
-## Eligibility rules
-
-Only benchmark targets whose status is currently `compiled`.
-
-Use:
-
-```bash
-interf status
-```
-
-or the benchmark wizard itself to confirm eligibility.
-
-If a target is `stale`, `failed`, or `idle`, do not benchmark it yet. Recompile first.
-
-## Recommended flow
-
-1. Confirm the source folder and target type.
-2. Confirm the selected workspaces are compiled.
-3. Reuse an existing benchmark spec if one already expresses the user's assertion.
-4. Otherwise create a small benchmark spec with:
-   - one clear question per case
-   - one output file per case
-   - a few deterministic checks
-5. Run `interf test`.
-6. Read the saved run artifact under `interf/benchmarks/runs/`.
-7. Summarize pass/fail differences and what they imply about the workflow or workspace.
-
-## Authoring guidance
-
-Prefer:
-- small JSON files
-- plain English
-- deterministic checks
-- reusable assertions that can be run again after changes
-
-Avoid:
-- giant benchmark files
-- exact-answer matching
-- judge-style prompt evaluation as the default
-- workflow-specific assumptions hardcoded into the benchmark unless the benchmark is intentionally for that workflow
-
-## Reporting guidance
-
-When summarizing a benchmark run:
-- say which targets were compared
-- say which cases passed or failed
-- point to the saved run file
-- explain what the failures imply about the current workflow or workspace
-- keep benchmark conclusions separate from subjective product opinions

package/skills/workflow/create/SKILL.md

@@ -1,141 +0,0 @@
----
-{
-  "name": "workflow/create",
-  "description": "Create or refine an Interf Compiler workflow package. Use when asked to define a workspace workflow, author workflow.json, shape compile stages, or add declarative stage acceptance criteria."
-}
----
-
-# Create Workflow
-
-Use this skill when shaping a reusable Interf Compiler workflow package.
-
-Prefer the CLI first:
-
-```bash
-interf create workflow
-```
-
-That wizard should create the initial package under:
-
-```text
-interf/workflows/<workflow-id>/
-```
-
-Then refine the package in place.
-
-## What a workflow is
-
-A workflow is a package, not just a JSON file.
-
-It has two layers:
-
-- `workflow.json` = machine-readable stage graph and contract mapping
-- `workspace.schema.json` = deterministic compiled-workspace shape
-- local docs = human/agent method layer
-
-Typical shape:
-
-```text
-interf/workflows/<workflow-id>/
-  workflow.json
-  workspace.schema.json
-  README.md
-  create/
-    SKILL.md
-  compile/
-    stages/
-      <stage-id>/
-        SKILL.md
-  use/
-    query/
-      SKILL.md
-```
-
-## What `workflow.json` should define
-
-Keep it declarative.
-
-- `id`
-- `type`
-- `label`
-- `hint`
-- optional `extends`
-- optional `stages`
-- optional `stage_policy_notes`
-
-Each stage can define:
-
-- `id`
-- `label`
-- `contract_type`
-- optional `skill_dir`
-- optional `description`
-- optional `acceptance`
-
-## Contract kinds
-
-Use the shipped workspace contract families only:
-
-- `workspace-file-evidence`
-- `workspace-knowledge-structure`
-- `workspace-query-shape`
-
-Do not invent new engine contract kinds in a local workflow package.
-
-## Workspace schema
-
-Every workflow package should also define `workspace.schema.json`.
-
-It should describe the deterministic output zones Interf Compiler expects, including:
-
-- `raw/`
-- `summaries/`
-- `knowledge/entities/`
-- `knowledge/claims/`
-- `knowledge/indexes/`
-- `home.md`
-- `.interf/`
-
-Use `owned_by` to show which stage ids are responsible for each zone. This keeps the reusable workflow package aligned with the compiled workspace ABI.
-
-## Declarative acceptance
-
-Use stage `acceptance` for deterministic stage checks that the CLI can compare against runtime state.
-
-Current acceptance fields:
-
-- `artifacts_exist`
-- `state_truthy`
-- `state_equals_counts`
-- `state_at_least`
-- `state_at_least_counts`
-
-Use dynamic counts when possible.
-
-Good:
-- `"summarized": "target_files"`
-- `"structured": "summary_total"`
-- `"shaped": "summary_total"`
-- `"compiled": "summary_total"`
-
-Avoid hardcoding fixed file counts unless the workflow really requires them.
-
-## Builder rules
-
-- keep `workflow.json` readable and declarative
-- put execution guidance in `SKILL.md` files, not in dense JSON prose
-- let the CLI own runtime state, verification, and pass/fail
-- use `AGENTS.md` as the workspace router after scaffolding, not as the reusable workflow spec
-- prefer extending shipped workflows before inventing a new stage graph
-
-## Outcome
-
-A good workflow package gives Interf Compiler:
-
-- a compile pipeline
-- a deterministic workspace schema
-- stage-to-contract mapping
-- readable local docs for executors
-- optional declarative acceptance checks
-
-The goal is not to write code inside the workflow. The goal is to define a stage pipeline that Interf Compiler can enforce.

package/skills/workspace/shape/SKILL.md

@@ -1,15 +0,0 @@
-# Workspace Shape
-
-Shape the final workspace surface around the workspace focus and saved truth checks.
-
-Requirements:
-- read the shaping input named in the stage contract
-- treat the shaping input as focus plus question text only
-- write `home.md` and any focused indexes needed under `knowledge/indexes/`
-- bias routing, prioritization, and navigation toward the job this workspace should answer well
-- use the existing knowledge structure instead of redoing extraction work
-- do not copy expected answers into the workspace
-- prefer better retrieval routes over speculative synthesis
-
-Output goal:
-- a final workspace surface that is easier for agents to navigate and use correctly

package/skills/workspace/structure/SKILL.md

@@ -1,15 +0,0 @@
-# Workspace Structure
-
-Build the cross-file knowledge structure from the summary set.
-
-Requirements:
-- read `summaries/` and `.interf/inventory.json`
-- write durable notes under `knowledge/entities/`, `knowledge/claims/`, and `knowledge/indexes/`
-- canonicalize entities, claims, categories, timelines, and other retrieval structure
-- use taxonomy and ontology only to improve retrieval, navigation, and evidence tracking
-- prefer small stable notes over one giant catch-all file
-- do not write `home.md` in this stage
-- do not copy expected answers from saved checks
-
-Output goal:
-- a reusable cross-file structure that later shaping can route through

package/skills/workspace/summarize/SKILL.md

@@ -1,15 +0,0 @@
-# Workspace Summarize
-
-Turn raw source files into per-file evidence notes under `summaries/`.
-
-Requirements:
-- process only the files listed in the stage contract
-- write one markdown summary per source file
-- use JSON frontmatter, not YAML
-- include `source`, `source_kind`, `evidence_tier`, `truth_mode`, `state`, and `abstract`
-- include a readable `## Abstract` block in the body
-- keep summaries source-grounded and conservative
-- do not do broad cross-file synthesis in this stage
-
-Output goal:
-- a clean per-file evidence layer the later stages can trust

package/templates/workspace/README.md

@@ -1,24 +0,0 @@
-# Workspace Template
-
-This directory contains bundled helper files for compiled workspaces.
-
-Interf Compiler creates compiled workspaces under:
-
-```text
-interf/workspaces/<name>/
-```
-
-The generated workspace is a file-based layer on top of your raw files.
-
-Interf Compiler scaffolds:
-
-- `interf.json`
-- `AGENTS.md`
-- `CLAUDE.md`
-- `home.md`
-- `workflow/`
-- `workflow/workspace.schema.json`
-- `.interf/`
-- `.interfignore`
-
-The source folder remains the human control plane for the dataset. The compiled workspace is the agent-facing surface.

package/templates/workspace/interfignore

@@ -1,2 +0,0 @@
-.obsidian/
-.DS_Store