build: extract shared boilerplate via generator (trust boundary/lang/pre-flight/reviewer hint)

[JFK]

Problem

~2,300 tokens of boilerplate are duplicated near-verbatim across 8-9 command files: the Trust-boundary forbidden-action list, the Output-language policy, the pre-flight validation bash, and the reviewer language hint. Slash commands have no runtime include mechanism, so this is primarily a maintainability / sync-bug problem (the jq filter already needs a CI check to stay in sync), not a pure runtime-token win.

Approach

Author the shared blocks once as fragments and add a build/generate step that inlines them into each commands/*.md at author time. Runtime files stay self-contained (no extra Read per invocation); the duplication-induced sync bugs disappear.

Scope

• Identify the canonical shared blocks: Trust-boundary clauses, Output-language policy, pre-flight checks, reviewer language hint (evidence: cross-file Explore review 2026-06-13).
• Add fragments/ + a generator (script) that composes command files from a template + fragments.
• Regenerate all command files; confirm byte-stable output vs hand-maintained versions (modulo the intended single-source change).
• CI check that generated files are up to date (fail if a command file drifts from its fragments) — this can retire the bespoke jq-sync check.

Acceptance criteria

• All 9 command files build from fragments; git diff after regenerate is empty on a clean tree.
• Editing a shared fragment updates every consumer in one place.
• Be explicit in the PR: runtime tokens are roughly unchanged; the win is maintainability + sync-bug elimination.

Depends on

#3, #4, #5, #6 (compress first, then extract the stabilized shared blocks). Architectural — lowest priority in the milestone.

[JFK]

Closing as a design decision, not an autonomous refactor (see #96).

The boilerplate duplication is real (~240-350 lines across 7-8 command files: Trust boundary, Output language, pre-flight bash, reviewer language hint). But #93 as written has three problems:

1. It does not reduce runtime tokens. The issue itself admits each generated file still inlines the block — the value is maintainability / sync-bug elimination, not token reduction.
2. It conflicts with an explicit design principle. CONTRIBUTING.md:24: "There is intentionally no scripts/ directory — every algorithm lives in the command markdown body." A fragment generator + CI "generated files up to date" check is exactly the infrastructure that principle excludes.
3. The "retire jq-sync-check" claim is incorrect. tests/jq-sync-check.sh enforces semantic equivalence between two deliberately-different formulations of the Copilot-detection filter — a text-inlining generator does not address that, so the check must stay.

If the maintainer wants to revisit the no-scripts principle for non-algorithmic boilerplate, that's a worthwhile design discussion — please reopen with that framing rather than as a token-reduction task.