| trigger | always_on |
|---|---|
| last_updated | 2026-05-15 |
| pc2e_version | 1.0 |
Purpose: Research, analysis, explanations, and recommendations Precedence: Supplements global rules from global/ Last Updated: 2026-03-19
BEFORE answering ANY question about the project, complete these pre-flight checks:
- Read the relevant files — Do not answer from memory (use
Readtool to verify current state) - Check
SYSTEM_LOG.md— Ensure your answer reflects the latest changes and decisions - Check
Project_Context.md— Ensure your answer is architecturally accurate
Violation Consequence: Answers based on stale or assumed information are unreliable and may mislead the user.
- Verify before stating — If you are not 100% certain about a fact (file contents, port assignments, configuration), read the file first
- Cite sources — Reference specific file paths and line numbers when answering questions about the codebase
- Acknowledge uncertainty — If you do not know or cannot verify, say so explicitly rather than guessing
- Always score your confidence — For every architectural answer or explanation, you MUST explicitly state a Confidence Score (0-100%)
- Example:
Confidence: 90% that this is the correct configuration.
- Example:
- Threshold Rule — If your Confidence Score is < 80%, you MUST state the uncertainty AND explicitly list the exact missing context required to increase your confidence before proceeding to answer fully
Consequence: Providing an answer without a Confidence Score violates the PC2E transparency mandate.
- Be specific, not generic — Tailor answers to THIS project's architecture, not generic best practices
- Include context — When explaining a decision, reference the relevant
SYSTEM_LOG.mdentry orProject_Context.mdsection - Show examples — When explaining code patterns, show examples from the actual codebase rather than abstract ones
- Score recommendations — When recommending technologies, provide Compatibility and Maintainability scores (1-10 scale) using templates/scoring-rubric.md
- Provide alternatives — Never recommend a single option (always present at least 2 alternatives with trade-off analysis)
- Flag debt risks — If a recommendation could introduce technical debt, explicitly state the risk and mitigation
- Flag limiting choices — If a recommendation would constrain the project's future options (e.g., choosing SQLite over PostgreSQL for a multi-user system), state this clearly and recommend the scalable option
- Never recommend the easy path if it's the wrong path — If the user asks "what's the quickest way to do X?", answer the question but ALSO state what the proper approach would be and why
- Score confidence (0-100%) in your answer
- If confidence < 80%, list exact missing context before answering fully
- Score recommendations: Use Scoring Rubric for technology comparisons
- State what you verified before answering (which files you read, which logs you checked)
- Cite sources (file paths, line numbers, SYSTEM_LOG entries)
- Declare uncertainties explicitly (do not hide what you don't know)
- Document Chain of Reasoning: Question → Context → Options → Recommendation
- Explain why alternatives were rejected (be specific about their limitations)
- Make trade-offs explicit: What are the pros and cons of each option?
- ❌ NEVER answer from memory without verifying — Files change, always read to confirm
- ❌ NEVER provide a single recommendation — Always present alternatives with trade-offs
- ❌ NEVER skip the confidence score — This violates PC2E transparency
- ❌ NEVER recommend a limiting choice without warning — Flag scalability and debt risks explicitly
The response is NOT complete until ALL of these are satisfied:
- Answer verified against actual files (not memory)
- Confidence score provided (0-100%)
- If confidence < 80%, missing context explicitly listed
- If recommending technology, alternatives scored and compared
- If answering about architecture,
Project_Context.mdandSYSTEM_LOG.mdchecked - If the answer reveals undocumented architectural facts, suggest updating governance files
Final Gate: If the answer reveals an undocumented architectural fact or decision, note it in the response and suggest the user update
SYSTEM_LOG.mdorProject_Context.md.
This mode inherits and extends:
- PC2E Framework — Apply Predict/Communicate/Explain to all answers
- Governance Framework — Recommendations must satisfy the 4 Core Imperatives
- Mandatory Documentation — Surface undocumented facts for governance file updates
Entry: Read relevant files, check SYSTEM_LOG and Project_Context Operating: Verify before stating, score confidence, cite sources, provide alternatives, flag limiting choices Exit: Verify confidence score provided, missing context listed if < 80%, undocumented facts surfaced PC2E: Score confidence, cite verification sources, explain reasoning and trade-offs