Agent Skills: Argument Snapshot

Argument Snapshot

Create the final C5 snapshot consumed by merge and later audit.

Despite the historical Skill name, the current implementation is not an autonomous rewrite loop. It reads the final H3 files, assigns bounded structural move labels, writes a compact consistency contract, refreshes section hashes, and fails when required prose is absent or too thin.

Position In The Workflow

section writing
-> style and numeric hygiene
-> logic polish
-> paragraph compaction
-> argument snapshot
-> optional transitions
-> merge

Running the snapshot after every section mutator prevents the ledger and manifest from describing an earlier draft.

Inputs

• sections/
• outline/outline.yml
• queries.md
• output/PARAGRAPH_CURATION_REPORT.md

Outputs

• output/ARGUMENT_SELFLOOP_TODO.md
• output/SECTION_ARGUMENT_SUMMARIES.jsonl
• output/ARGUMENT_SKELETON.md
• refreshed sections/sections_manifest.jsonl

SECTION_ARGUMENT_SUMMARIES.jsonl contains one record per expected H3 and one record per paragraph. Current move labels are deterministic signals drawn from:

setup, thesis, contrast, evidence, evaluation, limitation, synthesis, takeaway

They make section shape inspectable; they do not prove that an argument is scientifically valid. ARGUMENT_SKELETON.md contains a ## Consistency Contract and a chapter-level map. It is a writer-facing audit artifact, never reader-facing paper content.

PASS Contract

• every expected H3 exists;
• every H3 meets the profile paragraph floor;
• each paragraph has at least one allowed move;
• ARGUMENT_SKELETON.md contains ## Consistency Contract;
• sections_manifest.jsonl carries current bytes and sha256 values.

On FAIL, inspect ARGUMENT_SELFLOOP_TODO.md and route the named section back to its writer or evidence owner. This Skill does not apply the repair itself.

Run

uv run python .codex/skills/argument-selfloop/scripts/run.py \
  --workspace workspaces/<name>

Optional runner fields are --unit-id, --inputs, --outputs, and --checkpoint.