Conversation
…ctions)
MODULE_OVERVIEW_PROMPT and REPO_OVERVIEW_PROMPT (module-parent and repo-root
overview generation) have no {custom_instructions} slot, unlike
SYSTEM_PROMPT/LEAF_SYSTEM_PROMPT — so --instructions (language, audience,
forbidden diagram styles, link format) never reaches these two prompts.
Confirmed empirically on a real run (voll-intelligence): every affected
overview page came back in English with the tool's default developer-oriented
structure ("Purpose / Architecture / Core component documentation") and, in
one case, a broken local-filesystem link — regardless of a detailed
Portuguese business-audience --instructions file being passed on the CLI.
Appending the instructions as trailing text to the single user-role prompt
(the obvious first fix) measurably failed to change model output. The actual
fix: thread an optional system_prompt through call_llm/_call_llm_via_litellm/
_call_llm_via_azure/LLMBackend.complete/PydanticAIBackend.complete/
CawBackend.complete, and have generate_parent_module_docs pass
config.get_prompt_addition() as a system-role message — the same mechanism
SYSTEM_PROMPT/LEAF_SYSTEM_PROMPT already use successfully, just not
previously wired to the bare-completion overview path.
…gnature The previous commit added an optional system_prompt kwarg to LLMBackend.complete() (and a get_prompt_addition() call on the config passed into generate_parent_module_docs), but two test doubles didn't get updated to match: - FakeBackend.complete() in test_processing_order_update.py and test_updater_orchestrator.py didn't accept system_prompt, so generate_parent_module_docs's `backend.complete(prompt, system_prompt=...)` call raised TypeError. - The SimpleNamespace fake config in test_processing_order_update.py had no get_prompt_addition(), so the AttributeError below. CI: 3 failed (AttributeError: 'SimpleNamespace' object has no attribute 'get_prompt_addition'; TypeError: FakeBackend.complete() got an unexpected keyword argument 'system_prompt').
This branch has not been deployed
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Problem
MODULE_OVERVIEW_PROMPTandREPO_OVERVIEW_PROMPT(module-parent and repo-root overview generation) have no{custom_instructions}slot, unlikeSYSTEM_PROMPT/LEAF_SYSTEM_PROMPT— so--instructions(language, audience, forbidden diagram styles, link format) never reaches these two prompts.Confirmed empirically on a real run (
voll-intelligence, business documentation in Portuguese for a PM/PO audience): every affected overview page came back in English with the tool's default developer-oriented structure ("Purpose / Architecture / Core component documentation") and, in one case, a broken local-filesystem link (file:///home/.../docs/...) — regardless of a detailed Portuguese business-audience--instructionsfile being passed on the CLI.What didn't work
Appending the instructions as trailing text to the single user-role prompt (the obvious first fix) measurably failed to change model output — still English, still the default structure.
Fix
Thread an optional
system_promptthroughcall_llm/_call_llm_via_litellm/_call_llm_via_azure/LLMBackend.complete/PydanticAIBackend.complete/CawBackend.complete, and havegenerate_parent_module_docspassconfig.get_prompt_addition()as a system-role message. This is the same mechanismSYSTEM_PROMPT/LEAF_SYSTEM_PROMPTalready use successfully (via the Agent'ssystem_prompt=) — it just wasn't wired to the bare-completion overview path.Validation
Isolated test call with the fix (same model, same instructions, synthetic minimal
repo_structure) produced fully Portuguese, correctly-structured, correctly-hedged output. Applied to the realvoll-intelligencerun: all previously-broken overview pages (including the repo root) came back in the correct language/structure after the fix.🤖 Generated with Claude Code