feat(chat): add ON_PROMPT_READY plugin hooks + telemetry-based prompt middleware example

[mrveiss]

Summary

• Added ON_SYSTEM_PROMPT_READY and ON_FULL_PROMPT_READY hook points to the extension hook enum (hooks.py + base.py)
• Wired both hooks into llm_handler.py prompt construction pipeline — plugins can intercept and modify prompts at system-prompt and full-prompt stages
• Created plugins/core-plugins/telemetry-prompt-middleware/ — queries Prometheus CPU; appends concise-response hint when load exceeds threshold (default 80%)
• Added docs/developer/PROMPT_MIDDLEWARE_GUIDE.md — full reference for the new hook interface with worked examples
• 9 new tests in prompt_hooks_test.py covering enum presence, arg delivery, return-value replacement, None passthrough, and error isolation

Closes #3405

[github-actions]

✅ SSOT Configuration Compliance: Passing

🎉 No hardcoded values detected that have SSOT config equivalents!

[mrveiss]

Code review

Found 2 issues.

1. False "modified" debug log fires on every request when no prompt extension is registered (misleading log noise)

Both _emit_system_prompt_ready and _emit_full_prompt_ready call invoke_with_transform, which returns the original unchanged value when no extension is registered. The guard if isinstance(result, str) and result: evaluates to True for any non-empty prompt, so the function logs "[#3405] ON_SYSTEM_PROMPT_READY modified system prompt (N -> N chars)" even though nothing was modified. This fires on every chat request when no prompt extension is loaded.

AutoBot-AI/autobot-backend/chat_workflow/llm_handler.py

Lines 55 to 61 in 8672f1d

	if isinstance(result, str) and result:
	logger.debug(
	"[#3405] ON_SYSTEM_PROMPT_READY modified system prompt "
	"(%d -> %d chars)",
	len(system_prompt),
	len(result),
	)

The fix is to compare result is not context.get("system_prompt") (identity check after invoke_with_transform) or to check result != system_prompt before logging, not just truthy-string.

2. plugin.json hooks field declares the wrong method name — inconsistent with existing plugins and will mislead developers using this as a reference implementation

Every other plugin.json in the repo uses the hook method name (e.g., "on_agent_execute" for AGENT_EXECUTE). Because ON_FULL_PROMPT_READY gets the double-prefix treatment (on_ + on_full_prompt_ready), the correct method name is on_on_full_prompt_ready. The new telemetry plugin declares "hooks": ["on_full_prompt_ready"] — wrong by one on_ prefix. This is also documented incorrectly in PROMPT_MIDDLEWARE_GUIDE.md as a metadata-only issue today, but will break any future plugin loader that validates the field.

AutoBot-AI/plugins/core-plugins/telemetry-prompt-middleware/plugin.json

Line 24 in 8672f1d

"hooks": ["on_full_prompt_ready"]

The root cause is that ON_SYSTEM_PROMPT_READY and ON_FULL_PROMPT_READY enum members already contain the ON_ prefix, producing the awkward on_on_* method names. Consider renaming the enum members to SYSTEM_PROMPT_READY / FULL_PROMPT_READY so the dispatch produces the cleaner on_system_prompt_ready / on_full_prompt_ready — consistent with all other hook naming in the codebase.

---

🤖 Generated with Claude Code

_{- If this code review was useful, please react with 👍. Otherwise, react with 👎.}