Unify prompt suffix injection with ordered providers

[tninja]

Summary

• Add an ordered ai-code-prompt-suffix-functions provider pipeline with a shared prompt context and memoized classification.
• Migrate the custom suffix, Grill, auto-test/TDD, and discussion follow-up behaviors to providers in a deterministic order.
• Preserve Grill command origin, fail closed on provider errors, and remove legacy advice during hot reloads.

Why

Prompt suffixes were injected through a mix of inline logic and feature-specific advice. That made ordering implicit, coupled unrelated features, and made future suffix extensions harder to add safely.

Impact

Normal prompt sends now pass through one extensible suffix boundary. Single-token slash commands and raw resend paths continue to bypass suffix providers. The legacy ai-code-use-prompt-suffix switch controls the custom, auto-test, and follow-up providers without disabling Grill.

Root cause

Suffix ownership was distributed across ai-code-prompt-mode.el, Grill advice, and harness advice. Long-lived Emacs sessions could also retain the old advice after an upgrade, causing duplicate prompts or classification.

Validation

• Full ERT suite: 1059 tests, 1046 expected, 0 unexpected, 13 optional-dependency skips.
• Focused prompt-mode, Grill, and harness suite: 143/143 passed.
• Touched production files byte-compiled successfully with no new warnings.
• Checkdoc reported no new warnings in touched files.
• All six touched source/test files are clean versus the diagnostics baseline.
• git diff --check passed.
• Independent code review found no remaining Critical, Important, or Minor issues.

[Copilot]

Pull request overview

This PR refactors prompt suffix injection into a single, deterministic “provider pipeline” that runs at the send boundary, so feature-specific suffix behaviors (custom suffix, Grill, auto-test/TDD, discussion follow-up) no longer rely on scattered inline logic or long-lived advice.

Changes:

• Introduces ai-code-prompt-suffix-functions as an ordered provider hook, with a shared ai-code-prompt-context and memoized per-prompt classification.
• Migrates Grill and harness send-time suffix logic to providers (and removes legacy advice on reload to prevent duplicate behavior in long-lived sessions).
• Adds targeted ERT coverage for provider ordering, memoization semantics (including caching nil), fail-closed behavior, and reload/idempotency concerns.

Reviewed changes

Copilot reviewed 7 out of 7 changed files in this pull request and generated no comments.

Show a summary per file

File	Description
ai-code-prompt-mode.el	Adds the prompt context + ordered suffix provider pipeline and applies it at the ai-code--write-prompt-to-file-and-send boundary via :filter-args.
ai-code-grill.el	Replaces prompt-transform advice with a Grill suffix provider, while preserving “origin command” tracking via advice on entry commands.
ai-code-harness.el	Moves auto-test and discussion follow-up suffix routing into providers sharing memoized classification; removes legacy around-advice if present.
test/test_ai-code-prompt-mode.el	Adds tests for provider order, memoization caching of nil, switch-respecting custom provider behavior, and provider error abort semantics.
test/test_ai-code-harness.el	Adds integration-style tests for built-in provider ordering, reload cleanup of legacy advice, Grill independence from the legacy switch, and shared classification.
test/test_ai-code-grill.el	Updates Grill tests to validate provider-based behavior and reload cleanup, and keeps direct-command bypass verified via insertion path.
docs/superpowers/plans/2026-07-15-prompt-suffix-providers.md	Adds an implementation plan document capturing the intended architecture and verification steps.