docs: remove em-dashes and AI-tell wording from markdown prose

[chaliy]

What changed

Removes em-dashes from markdown prose across the repo (72 files: docs/, specs/, crates/bashkit/docs/, per-crate READMEs, examples, bench results, site apidocs, top-level README/AGENTS) and replaces them with plain punctuation: colons for label/definition lists, semicolons or periods for clauses an em-dash was joining, parentheses for appositive lists. Also rewords one AI-tell word (seamlessly to directly). Documentation prose only.

Why

Em-dashes read as an "AI-written" tell. This is the bashkit slice of an ecosystem-wide prose cleanup. No behavior change.

Before / After

Documentation-only, no observable behavior change. The doc-tested embedded guides under crates/bashkit/docs/*.md had only prose (outside all code fences) edited; the fenced rust examples that are compiled/doc-tested were left byte-identical.

Risk

• Low
• What can break: nothing functional; only markdown prose changed; no code, fixtures, or snapshots touched. Em-dashes inside code fences were deliberately preserved.

Checklist

• Tests added or updated (n/a, docs only)
• Backward compatibility considered (no API/behavior change)

[cloudflare-workers-and-pages]

Deploying with    Cloudflare Workers

The latest updates on your project. Learn more about integrating Git with Workers.

Status	Name	Latest Commit	Preview URL	Updated (UTC)
✅ Deployment successful! View logs	bashkit	9c36b31	Commit Preview URL Branch Preview URL	Aug 17 2026, 05:16 PM

[chaliy]

CI is red here. Root cause, verified against the job log and the generator source:

Failing job: Lint (job 95120566879) → python3 -m unittest discover -s scripts/tests → test_capability_parity.CapabilityParityContractTests.test_generated_inventory_is_current. The Check job is just the aggregate gate and fails only because Lint did — one real failure, not two.

Why: the sweep edited knowledge/status/capability-parity.md, which is a generated file, not prose. In that table — is the semantic "unsupported" marker, emitted by scripts/capability_parity.py:154:

"✅" if capability_id in surfaces[surface_id]["supported"] else "—"

Stripping the em-dashes left the unsupported cells empty, so the committed file no longer matches render_inventory(load_manifest()). That's exactly what the assertion diff shows — | ✅ | — | — | — | in the expected output vs | ✅ | | | | in the committed one. The See also lines at scripts/capability_parity.py:178-180 emit em-dashes too and have the same problem.

Fix: restore that file to generator output and keep it out of the prose sweep:

just regen-capability-parity   # python3 scripts/capability_parity.py

Worth excluding generated outputs from the em-dash pass generally — knowledge/status/ is generated from contracts/capability-parity.json, so any hand-edit there will drift the same way. The other 86 files in this PR look unaffected; all 43 remaining checks are green.

Not pushing the fix myself — this PR is on claude/peaceful-hawking-5hqjw0 and I'm scoped to a different branch.

---

Generated by Claude Code