[instructions] Sync instruction files with release v0.79.8

[github-actions]

Summary

Syncs three instruction files under .github/aw/ with the v0.79.8 release. The changes update experiment metric naming, deprecate a legacy workflow-syntax alias, and document per-token pricing for custom/private models — no production code is touched.

Changes

.github/aw/experiments.md (modified)

Renames the A/B test metric from effective_tokens to ai_credits and updates the corresponding hypothesis text to match. Keeps the verbosity experiment configuration otherwise intact.

.github/aw/syntax-agentic.md (modified)

Demotes max-runs: from a primary top-level field to a deprecated legacy alias for max-turns:. max-turns: is now documented first; a migration hint pointing to gh aw fix is added alongside max-runs:.

.github/aw/token-optimization.md (modified)

Adds a paragraph documenting the top-level models: frontmatter field. This field supplies per-token pricing (input / output / cache_read / cache_write) for custom or private models and is merged with the built-in models.json at runtime.

Impact

File	Impact	Breaking
experiments.md	medium — metric rename affects any tooling that reads experiment config	No
syntax-agentic.md	medium — max-runs: is now explicitly deprecated; callers should migrate	No
token-optimization.md	low — documentation-only addition	No

Notes

• All changes are confined to .github/aw/ instruction/documentation files; no Go source, schemas, or test files are modified.
• The max-runs: → max-turns: deprecation is surfaced via gh aw fix (codemod), so existing workflows continue to function without an immediate change.
• The ai_credits metric rename in experiments.md should be cross-checked against any dashboard or observability queries that previously referenced effective_tokens.

Generated by PR Description Updater for issue #39043 · 97.5 AIC · ⌖ 12.5 AIC · ⊞ 19.8K · ◷

[github-actions]

Hey @github-actions 👋 — great work from the Instructions Janitor on keeping the .github/aw/ instruction files in sync with release v0.79.8! The three targeted fixes (deprecating max-runs in favour of max-turns, documenting the new models: frontmatter field, and renaming effective_tokens → ai_credits) are accurate, well-scoped, and backed by a thorough audit trail.

One flag from the automated checklist worth noting:

• No test files changed — The checklist requires test coverage for a lgtm verdict. For this PR that's a mechanical false positive: all three changed files are Markdown instruction files under .github/aw/, and there are no applicable unit or integration tests for them. If a linting or snapshot-based test suite for instruction-file content exists (e.g. checking line counts or known-deprecated terms), wiring the janitor output through it would let future syncs clear this check automatically.

If you'd like to address the test gap, you can assign this prompt to a coding agent:

Check whether any existing test or lint suite in this repo validates .github/aw/*.md instruction files (e.g. line-count limits, banned deprecated terms, required section headings). If such a suite exists, ensure the three files modified in PR #39043 pass it. If no such suite exists, create a minimal Go test (or shell-based CI check) under scripts/ or internal/ that:
1. Asserts each .github/aw/*.md file is ≤ 500 lines.
2. Asserts the string "effective_tokens" does not appear in any .github/aw/*.md file (replaced by ai_credits).
3. Asserts the string "max-runs" only appears in a deprecated/legacy context (i.e. is preceded by "deprecated" within the same paragraph) in syntax-agentic.md.
Add the new check to the Makefile lint or test target so it runs in CI.

Generated by ✅ Contribution Check · 363.1 AIC · ⌖ 13.4 AIC · ⊞ 24.7K · ◷