fix(skills): use disk extraction prompts for write-capable hosts

[rajanpanth]

Summary

• generate verbose extraction guidance for Claw, Kiro, and Pi because their dispatched agents can write chunk files
• retain Codex's compact inline-only prompt
• lock the dispatch/prompt contract in skill generator tests

Fixes #2483

Validation

• uv run python -m tools.skillgen --check
• uv run pytest tests/test_skillgen.py -q

[graphify-labs]

Graphify reviewed this change.

Worth a look — the grounded gate found no coupling regressions or blocking issues, but 1 advisory finding(s) below merit a look before merge.

---

Graphify review — findings

This pull request replaces the "compact" version of the extraction subagent prompt in extraction-spec.md with a more verbose, expanded version, affecting the claw and kiro skill reference files (and presumably others per the truncation). The changes add a new CHUNK_PATH substitution variable and instructions to write JSON output to disk, expand the rules with detailed examples for image types, confidence scoring rubrics, node ID formatting, and source_file handling. A large set of test_skillgen tests and expected-output fixtures are also updated to match the new prompt content.

Worth a look

• Extraction prompt changes result delivery from returned JSON to mandatory file write — graphify/skills/claw/references/extraction-spec.md:71 · Escalate · high
  • agreed by 2 of 2 members but NOT verified (no proof, no reproducing execution) — consensus is not a verdict; needs human review Execution auto-disposal is off for this run; enable it (with sandbox isolation) to have Graphify try to confirm or refute this automatically.

Analysis details — impact, health, verification

Impact & health

Graphify review

Impact — 146 functions depend on the 146 functions this change touches.

Health — grade A; 1 existing hotspot(s) in the area this change touches (pre-existing, not introduced here):

• test_audit_catches_a_dropped_non_allowlisted_heading() — 0 callers, 6 callees (medium)

Verification — 146 functions in the blast radius were not formally verified this run (proofs are advisory here).

Gate & verification

graphify gate

PASS — objectively clean (no health regressions, tests not run — proofs not run this pass (advisory)). Grounded, not self-assessed.

Advisory (not blocking):

• verification_scope: 146 function(s) in the blast radius were not formally verified this run