Ports Quarto 1's project.pre-render / project.post-render script support to Q2, for all project types. Strand: bd-w348iu63; plan: claude-notes/plans/2026-07-29-pre-post-render-scripts.md.

What

project:
  type: website
  pre-render: prepare.py          # string or list; full command lines OK
  post-render:
    - cleanup.R
    - python3 tools/notify.py --label "site build"

• Runner (quarto-core::project::render_scripts): string-or-list config extraction with YAML source locations; Q1-parity quote-aware command-line parser; dispatch on the first token's extension (.py → QUARTO_PYTHON/python3, .r → knitr find_rscript, .ts/.js → QUARTO_NODE/node, anything else direct-exec resolved against the project root); the QUARTO_PROJECT_* env contract; QUARTO_USE_FILE_FOR_PROJECT_{INPUT,OUTPUT}_FILES escape hatch (Q1 #10828). Exec half is native-gated, extraction compiles on WASM.
• q2 render: discover → pre-render → re-discover (script-created inputs render in the same pass) → mutation guard (project.type / project.output-dir frozen, Q1-compatible) → pipeline → post-render with env computed fresh from actual outputs (fixes Q1's stale-env wart). New --no-render-scripts flag.
• q2 publish: same bracket; post-render runs before the sidecar walk so script-added output files get published.
• q2 preview: pre-render scripts run once at boot (before eager captures), never re-run — a decided deviation from Q1's every-re-render behavior; failures keep the preview serving. No post-render in preview.
• Hub/WASM: scripts can't run in the browser; a one-time Q-5-12 warning is surfaced through the existing diagnostics channel.
• Diagnostics: Q-5-8 (script failed, ariadne snippet pointing at the exact pre-render: YAML scalar + exit code), Q-5-9 (forbidden mutation), Q-5-10 (launch failure), Q-5-11 (pre_render underscore-typo warning), Q-5-12 (unsupported environment).
• Docs: docs/guides/projects/scripts.qmd — config keys, env-var table, dispatch rules, preview behavior, Q1 differences.

Q1 → Q2 differences (deliberate)

Testing

TDD (all new tests verified failing before implementation):

• 14 CLI e2e tests (crates/quarto/tests/integration/render_scripts_cli.rs) driving the real q2 binary: script-created inputs, env contract full vs subset, failure abort semantics, mutation guards, ordering, quoted args, escape hatch, --no-render-scripts, direct-exec (#[cfg(unix)] sh / #[cfg(windows)] bat), typo warning. Python fixtures skip gracefully when no interpreter is present.
• 17 unit tests (parser, extraction, typo guard, mutation guard, catalog registration).
• 1 preview boot test (runs once at boot, does not re-run on file change).

cargo nextest run --workspace: 10806 passed. cargo xtask verify (full, incl. WASM leg): green. cargo xtask lint: clean. End-to-end verified manually per CLAUDE.md (fixture output inspected; recorded in the plan doc).

🤖 Generated with Claude Code