Audit findings: duplicated ~120-line skill boilerplate (51/56), over-prescriptive STRICT recipes, hardcoded 'Claude Opus 4.7' commit trailer

While auditing my local instruction files against Anthropic's Fable 5 prompting guidance, I ran a pass over the gstack skill suite (v1.60.1.0 as installed). Three findings, bundled here since they're all low-urgency maintenance items — happy to split them into separate issues if you prefer.

---

## Issue 1: Duplicated ~120-line boilerplate block across 51/56 skills

**Title:** 51/56 skills embed an identical ~120-line AskUserQuestion/Conductor/plan-tune boilerplate block

**Summary:**
An audit of the gstack skill suite found that 51 of 56 skill files contain a near-identical
~120-line boilerplate block covering AskUserQuestion setup, Conductor wiring, and plan-tune
integration. The duplication inflates every skill file and means any fix or update to this
shared logic has to be hand-propagated across 51 files.

**Details:**
- Example instance: `review/SKILL.md:335-457`.
- The block recurs near-verbatim in the remaining 50 skills that include it.
- Total duplicated content is roughly 51 × 120 = ~6,100 lines of repeated text across the suite.

**Suggested fix:**
Extract the shared block to a single include file (e.g. `_shared/askuserquestion-conductor.md`)
and have each skill reference it, or trim the boilerplate to the minimum each skill actually
needs. Either approach removes the multi-file edit burden and shrinks per-skill file size.

---

## Issue 2: Heavy STRICT step recipes degrade output on newer models

**Title:** Prescriptive STRICT phase/step recipes conflict with Anthropic's Fable 5 prompting guidance

**Summary:**
Several core skills (`spec`, `ship`, `autoplan`) use heavily prescriptive "STRICT" process
blocks with large numbers of MUST/step markers. Anthropic's guidance for Fable 5 notes that
instructions written for older, less capable models are "often too prescriptive ... and can
degrade output quality" on newer models — the STRICT recipes in this suite are exactly that
pattern.

**Details:**
- `spec/SKILL.md:847`: `Process (STRICT — do not skip or combine phases)`, followed by 92
MUST markers and 96 step markers within that skill alone.
- `ship/SKILL.md`: 71 MUST markers, 99 step markers.
- `autoplan/SKILL.md`: 67 MUST markers, 100 step markers.
- Suite-wide: 84 STRICT markers across the skill set.

**Suggested fix:**
Rewrite the STRICT recipes as intent + boundary statements (what must be true at the end,
what must not happen) rather than enumerated MUST/step sequences, per Anthropic's current
prompting guidance for capable models. This should reduce both token overhead and the
degradation risk the guidance describes.

---

## Issue 3: Hardcoded model name in commit trailers

**Title:** Commit trailers hardcode `Claude Opus 4.7`, will drift as models change

**Summary:**
Two skills generate git commit trailers with a hardcoded model name (`Claude Opus 4.7`)
instead of reflecting whichever model is actually active in the session. This trailer will
silently become inaccurate the moment a session runs on a different model.

**Details:**
- `document-generate/SKILL.md:1197`: `Co-Authored-By: Claude Opus 4.7`
- `ship/SKILL.md:1232`: `Co-Authored-By: Claude Opus 4.7`

**Suggested fix:**
Parameterize the trailer to the active model identity (if available to the skill at run time),
or fall back to a version-agnostic `Co-Authored-By: Claude` so the trailer doesn't need to be
updated by hand every time the underlying model changes.

Start with review/SKILL.md:335-457 and compare the repeated block across the other skill files, then inspect spec/SKILL.md:847, ship/SKILL.md, autoplan/SKILL.md, and the trailer lines in document-generate/SKILL.md:1197 and ship/SKILL.md:1232. Determine the intended shared structure and model identity behavior before changing files. Done means the duplication and prescriptive recipes are addressed consistently and trailers no longer drift from the active model.

Written by the indexing model from the issue text.