| name | Strict Adherence and Ingestion Order Protocol |
| description | MANDATORY for every workflow. Prevents context collapse, procedural fatigue, and 'Legacy Anchoring'. Forces the reading of agent instructions BEFORE user code. |
Strict Adherence and Ingestion Order Protocol
This Protocol is the architectural barrier against "Associative Complacency" and "Legacy Anchoring." Your model statistically tends to prioritize pre-existing formats and densities in the user's documents over your active mathematical instructions, corrupting the final result.
1. Unbreakable Sequence of Ingestion
When you receive a command to execute an algorithm, workflow, or regulated task (e.g., /derive, /test, /document), you MUST strictly comply with this temporal reading sequence. You are FORBIDDEN from reversing the hierarchy:
- Phase Alpha - Framework Ingestion: Use your tools to locate and read all relevant instructions stored within the
.agent/ directory (workflows, skills, templates associated with the task). You must saturate your context window with the PROCEDURAL RIGOR before contaminating yourself with the problem domain.
- Phase Beta - Tabula Rasa Purge: Proactively eliminate temporary artifacts or trash from your workspace (for example, by cleaning
.agent/scratch/ or discarding residual files in your environment) generated by previous executions of the same command.
- Phase Gamma - Environment Diagnostic (Only if Rule 04 is active): Execute the hygiene leveling (
MASTER-SPEC, TODO.md placeholders) if they are missing.
- Phase Delta - Contaminated Reading: ONLY AFTER assimilating the algorithm are you authorized to invoke tools to peek into and read the user's concrete project files (code, living documents, database).
2. Cognitive Isolation vs. Additive Expansion
When entering Phase Delta to read pre-existing specification documents (e.g., MASTER-SPEC.md §8), you must formatively evaluate your approach:
- Structural Amnesia: If the previous documents lack the canonical taxonomy or are in an obsolete format, apply "Categorical Blindness." Extract the empirical value but ruthlessly destroy and replace the format in your output.
- Additive Expansion (Nuance): If the previous documents already possess canonically valid records (e.g., checks like
[ACTOR.CAT.NN]), you MUST read them with extreme attention. Your mission in analytical workflows (like /derive) is not to destroy them, but to generate purely ADDITIONAL AND NON-OVERLAPPING nodes (MECE).
The "No-Op" Trap: If the file already has 18 perfectly formatted checks, and your mathematical algorithm detects actor cross-references that derive 140 new checks... YOU MUST NEVER BE LASSITUDINOUS. The fact that the document "already looks full" is no excuse to halt the algorithm. Accumulate and interleave the new mathematical density while respecting what already exists. Ignoring this is a Critical Execution Failure.
3. Workflow Anti-Skip (Procedural Fatigue Override)
You are ABSOLUTELY FORBIDDEN to:
- Infer Phases 1 through (N-1) "mentally" in order to swiftly deliver Phase N.
- Invent outputs (ghost data) without having traversed the empirical procedure of extracting their traceability step-by-step in the requested flow.
- Resolve gate checkpoints (GATES) without having previously established your strict reflection. Generative lethargy is not an excuse.
