compress-input

name	compress-input
description	Caveman-compress one policy doc into policy_facets/compressed/

Caveman-Compress One Policy Doc

Read one policy doc under <domain>/input/policy_docs/, copy it to its mirrored destination under <domain>/policy_facets/compressed/, and run /caveman:compress on the destination so the compressed mirror reflects the source.

This skill is the per-file primitive that mirrors /extract-computations's shape. It is invoked per file by /index-inputs's unified per-file batch worker (alongside /extract-computations for the same source file in a single subagent context) and may also be invoked standalone by the analyst against a single source file. The non-AI half (file enumeration, manifest sync, mirror-deletes, plan/finalize handoff for the whole domain) lives in xlator compress-inputs <domain> --plan and --finalize, called by /index-inputs. This skill is the AI half that does the per-file caveman-compression.

Input

/compress-input <path_to_policy_file>

<path_to_policy_file> must resolve to a path under <DOMAINS_DIR>/<domain>/input/policy_docs/. If not, pre-flight emits :::error and stops.

Read ../../core/output-fencing.md now.

Pre-flight

Run these checks before doing anything else. Skip pre-flight when invoked from inside /index-inputs's subagent worker — the parent already validated the path.

1. Argument provided?

   • NO → :::error Usage: /compress-input <path_to_policy_file> ::: Then stop.

2. Path resolves under <DOMAINS_DIR>/<domain>/input/policy_docs/?

   • Resolve <path_to_policy_file> to an absolute path. Accept absolute paths, paths relative to the project root, or paths relative to $DOMAINS_DIR.
   • Walk the resolved path's ancestors to find input/policy_docs/. The directory three levels up from input/policy_docs/<rel> (i.e., the ancestor whose child input/policy_docs/ exists) is <domain>. The grandparent is $DOMAINS_DIR.
   • If the resolved path is not under <DOMAINS_DIR>/<domain>/input/policy_docs/<rel> for any valid <domain> → :::error Path must be under /input/policy_docs/. Got: ::: Then stop.

3. Source file exists and is readable?

   • File missing or unreadable → :::error Source file not found: ::: Then stop.

4. Source has .md extension?

   • Non-.md → :::error Only .md sources are supported in v1; got: ::: Then stop.

No md_quality gate here. Compression is not quality-sensitive — /index-inputs enforces the quality gate at the orchestrator level by skipping REJECTED files before dispatching workers.

Process Checklist

This skill has 3 steps:

• Step 1: Compute destination and copy source
• Step 2: Invoke /caveman:compress on the destination
• Step 3: Print summary

Step 1: Compute destination and copy source

Compute the destination path:

<DOMAINS_DIR>/<domain>/policy_facets/compressed/<rel>.md

where <rel>.md is the source path relative to input/policy_docs/.

Create intermediate directories under policy_facets/compressed/ as needed and copy the source to the destination. Overwrite the destination unconditionally — the copy is idempotent. When invoked from /index-inputs, the destination already exists (the orchestrator's xlator compress-inputs --plan pre-copied it); the overwrite is a no-op in content terms but ensures the standalone-invocation path works without depending on --plan having run.

Step 2: Invoke /caveman:compress on the destination

Invoke:

/caveman:compress <abs path to dst under policy_facets/compressed/<rel>.md>

The caveman skill replaces the destination's contents with a token-reduced (caveman-format) version. It also creates a sibling <rel>.md.original.md backup; that backup is swept by xlator compress-inputs --finalize (or by xlator compress-inputs --plan's defensive sweep on the next run if the analyst invokes this skill standalone).

If /caveman:compress emits :::error, propagate the error and stop. Do not retry.

Step 3: Print summary

:::important ✓ Compressed: policy_facets/compressed/.md :::

Do NOT emit :::next_step from this skill — it is per-file and is normally invoked from a parent loop or subagent worker. The parent (e.g., /index-inputs) emits the workflow's next-step suggestion.

Common Mistakes to Avoid

• Don't update the manifest from this skill — the manifest is the single responsibility of xlator compress-inputs --finalize. When invoked standalone (outside /index-inputs), the destination file is overwritten with the compressed version but the manifest is not updated; the next xlator compress-inputs --plan run (typically via /index-inputs) will simply re-classify this file as to_compress because it has no manifest entry yet (or its SHA mismatches). This is acceptable best-effort behavior for the standalone path and matches /extract-computations's convention.
• Don't emit :::next_step — this skill runs inside a parent worker loop and the parent owns the next-step signal.
• Don't skip the source-to-dst copy — even when the destination exists from a prior --plan's pre-copy, an idempotent overwrite is required so the standalone-invocation path works without --plan.
• Don't run this skill on a non-.md source — pre-flight refuses; only .md is supported in v1.