Move RCA behavioral prompts to system prompt

[OlivierTrudeau]

Moves all investigation guidance out of the user message and into dedicated system prompt sections.

Why: The user message was ~130 lines of behavioral instructions mixed in with the actual alert data. The model treats system prompt as authoritative and non-truncatable, so methodology belongs there. The user message should just be the facts.

What changed:

• New rca_sections/ directory with 6 focused prompt files (identity, investigation, context_mgmt, error_recovery, evidence_standard, conclusion_gate)
• composer.py loads these for background RCA via _build_rca_system_prompt()
• rca_prompt_builder.py now returns only alert data + providers + Aurora Learn context — no inline instructions
• Deleted the old skills/rca/segments/ files that were being appended to the user message

Summary by CodeRabbit

• New Features

  • Added expanded investigation guidance: hypothesis-first workflow, explicit evidence standards, error-recovery steps, context recording, and a “before concluding” checklist to avoid premature root-cause claims.
  • Strengthened safeguards against common RCA mistakes (correlation vs causation, insufficient evidence, repeated failing queries).

• Refactor

  • Consolidated behavioral guidance into a single system-level prompt structure and simplified the prompt builder for clearer, less redundant investigator instructions.

[coderabbitai]

Warning

Review limit reached

@OlivierTrudeau, we couldn't start this review because you've reached your PR review rate limit.

More reviews will be available in 24 minutes and 39 seconds. Learn how PR review limits work.

Your organization has run out of usage credits. Purchase more in the billing tab.

⌛ How to resolve this issue?

After more reviews become available, a review can be triggered using the @coderabbitai review command as a PR comment. Alternatively, push new commits to this PR.

We recommend that you space out your commits to avoid hitting the rate limit.

🚦 How do rate limits work?

CodeRabbit enforces hourly rate limits for each developer per organization.

Our paid plans include higher PR review limits than trial, open-source, and free plans. In all cases, reviews become available again over time. During sustained high-volume PR review activity, CodeRabbit may temporarily slow when the next review becomes available.

Please see our Fair Usage Limits Policy for further information.

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: ASSERTIVE

Plan: Pro

Run ID: 89fdbf88-9c53-4aba-afd1-826994448af8

📥 Commits

Reviewing files that changed from the base of the PR and between 1eee810 and bf5a7af.

📒 Files selected for processing (4)

• server/chat/backend/agent/prompt/composer.py
• server/chat/backend/agent/prompt/rca_sections/conclusion_gate.md
• server/chat/backend/agent/prompt/rca_sections/identity.md
• server/chat/backend/agent/prompt/rca_sections/investigation.md

Walkthrough

Consolidates RCA behavioral guidance into ordered system-prompt markdown sections and updates background RCA assembly to load them. Removes legacy segment helpers and stops appending investigation guidance into the user message; builder now formats alert + context only.

Changes

RCA Prompt Architecture Restructuring

Layer / File(s)	Summary
RCA Behavioral Guidance Sections server/chat/backend/agent/prompt/rca_sections/identity.md, server/chat/backend/agent/prompt/rca_sections/investigation.md, server/chat/backend/agent/prompt/rca_sections/error_recovery.md, server/chat/backend/agent/prompt/rca_sections/context_mgmt.md, server/chat/backend/agent/prompt/rca_sections/evidence_standard.md, server/chat/backend/agent/prompt/rca_sections/conclusion_gate.md	Six ordered markdown sections define Aurora's identity, investigation workflow, error-recovery procedures, findings recording, evidence citation standards, and conclusion checklist for RCA system prompts.
System Prompt Assembly for Background RCA server/chat/backend/agent/prompt/composer.py	Adds _RCA_SECTION_ORDER and _build_rca_system_prompt(); updates build_system_invariant() so background RCA returns the concatenated RCA sections instead of the prior trimmed core-prompt approach.
RCA Prompt Builder Simplification server/chat/background/rca_prompt_builder.py, server/chat/backend/agent/skills/rca/segments/error_resilience_outro.md	Removes private RCA segment loading/rendering/appending helpers and eliminates appending investigation-depth/error-resilience/output requirements to the user message; the builder now emits only alert + context in the user message. A single-line skill segment was removed from error_resilience_outro.md.

Sequence Diagram

sequenceDiagram
  participant Composer
  participant RCAPrompt
  participant RCABuilder
  Composer->>RCAPrompt: _build_rca_system_prompt() reads ordered sections
  RCAPrompt-->>Composer: concatenated RCA system prompt
  Composer->>RCABuilder: build_system_invariant() returns system prompt
  RCABuilder->>RCABuilder: build_rca_prompt() formats alert + context (user message)
  RCABuilder-->>RCAPrompt: final RCA prompt (system prompt + user message)

Loading

🎯 3 (Moderate) | ⏱️ ~20 minutes

Possibly related PRs

• Arvo-AI/aurora#257: Related work on skills/prompt modularization and RCA segment/template loading.
• Arvo-AI/aurora#276: Overlaps on build_system_invariant() changes for background RCA prompt composition.
• Arvo-AI/aurora#411: Related edits to RCA investigation guidance content in skills segments.

Suggested reviewers

• damianloch
• beng360
• isiddharthsingh

Poem

🐰 In quiet burrows I arrange each line,

Sections ordered, stitched, and neatly fine.
System holds the guidance, user gives the fact,
Together they guide the investigation's tract.
Hops of joy — RCA in tidy design.

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title clearly and concisely summarizes the main architectural change: moving RCA behavioral prompts from the user message to the system prompt.
Docstring Coverage	✅ Passed	Docstring coverage is 100.00% which is sufficient. The required threshold is 80.00%.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch rca-agent-v2

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

[coderabbitai]

Actionable comments posted: 2

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@server/chat/backend/agent/prompt/composer.py`:
- Around line 41-47: The _build_rca_system_prompt function currently opens files
from _RCA_SECTIONS_DIR based on _RCA_SECTION_ORDER without guarding against
missing/unreadable files; update it to check existence (os.path.exists or
os.path.isfile) for each path before opening, wrap the open/read in a try/except
OSError block, log the error via the app logger or module logger (e.g.
app.logger.error or logging.getLogger(__name__).error) including the path and
exception, skip the missing/unreadable section and continue assembling parts,
and ensure the function returns a safe fallback (empty string or partial
assembled prompt) instead of letting the exception bubble up to Flask error
handlers.

In `@server/chat/backend/agent/prompt/rca_sections/identity.md`:
- Line 1: Add a top-level H1 heading at the start of identity.md (line 1) to
satisfy MD041 and match other RCA sections; prepend a descriptive H1 such as "#
Identity" (or "# Aurora" if you prefer agent-centric title) so the file begins
with an H1 instead of plain text.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: ASSERTIVE

Plan: Pro

Run ID: b5209aa5-9cbe-434e-b114-95f5e18aea58

📥 Commits

Reviewing files that changed from the base of the PR and between 916a8de and 6382d08.

📒 Files selected for processing (16)

• docs/RCA_QUALITY_ROADMAP.md
• server/chat/backend/agent/prompt/composer.py
• server/chat/backend/agent/prompt/rca_sections/conclusion_gate.md
• server/chat/backend/agent/prompt/rca_sections/context_mgmt.md
• server/chat/backend/agent/prompt/rca_sections/error_recovery.md
• server/chat/backend/agent/prompt/rca_sections/evidence_standard.md
• server/chat/backend/agent/prompt/rca_sections/identity.md
• server/chat/backend/agent/prompt/rca_sections/investigation.md
• server/chat/backend/agent/skills/rca/segments/critical_requirements_header.md
• server/chat/backend/agent/skills/rca/segments/error_resilience_infra.md
• server/chat/backend/agent/skills/rca/segments/error_resilience_intro.md
• server/chat/backend/agent/skills/rca/segments/error_resilience_outro.md
• server/chat/backend/agent/skills/rca/segments/output_requirements.md
• server/chat/backend/agent/skills/rca/segments/persistence_and_immediate_action.md
• server/chat/backend/agent/skills/rca/segments/what_to_investigate.md
• server/chat/background/rca_prompt_builder.py

💤 Files with no reviewable changes (7)

• server/chat/backend/agent/skills/rca/segments/critical_requirements_header.md
• server/chat/backend/agent/skills/rca/segments/error_resilience_outro.md
• server/chat/backend/agent/skills/rca/segments/error_resilience_infra.md
• server/chat/backend/agent/skills/rca/segments/what_to_investigate.md
• server/chat/backend/agent/skills/rca/segments/error_resilience_intro.md
• server/chat/backend/agent/skills/rca/segments/output_requirements.md
• server/chat/backend/agent/skills/rca/segments/persistence_and_immediate_action.md

[sonarqubecloud]

Quality Gate passed

Issues
1 New issue
0 Accepted issues

Measures
0 Security Hotspots
0.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud