From dbf1e75aa98dde99233f71207b74291a657ec37f Mon Sep 17 00:00:00 2001
From: Shahovalov MIkhail <m.shahovalov@yandex.ru>
Date: Mon, 13 Apr 2026 09:35:34 +0300
Subject: [PATCH] =?UTF-8?q?=D0=A3=D0=B4=D0=B0=D0=BB=D0=B5=D0=BD=D0=B8?=
 =?UTF-8?q?=D0=B5=20=D0=B2=D1=81=D0=BF=D0=BE=D0=BC=D0=BE=D0=B3=D0=B0=D1=82?=
 =?UTF-8?q?=D0=B5=D0=BB=D1=8C=D0=BD=D1=8B=D1=85=20=D0=B0=D0=B3=D0=B5=D0=BD?=
 =?UTF-8?q?=D1=82=D0=B0=20=D1=84=D0=B0=D0=B9=D0=BB=D0=BE=D0=B2?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

---
 AGENT.common.md      |  34 -------
 AGENT.interaction.md | 226 -------------------------------------------
 2 files changed, 260 deletions(-)
 delete mode 100644 AGENT.common.md
 delete mode 100644 AGENT.interaction.md

diff --git a/AGENT.common.md b/AGENT.common.md
deleted file mode 100644
index 87b8b5e..0000000
--- a/AGENT.common.md
+++ /dev/null
@@ -1,34 +0,0 @@
-# AGENT Common Rules
-
-Reusable baseline rules that are not tied to this repository.
-
-## Communication (MUST)
-- Answer first, then request approval if needed.
-- Concise, meaningful, no filler.
-- Do not end response with only procedural choice.
-- Secrets/config: use `.env.local` if present; if missing but required — state once, reference `.env.example` for keys; never commit `.env.local` or print secret values.
-
-Approval/clarification:
-- ask once, no repetition
-- binary -> yes/no
-- multiple -> numbered options + brief context
-- ask dependent questions sequentially, not in one message
-- if one decision changes or gates the next step, ask the prerequisite decision first and wait for the answer before asking the next one
-- do not combine branch selection with operational approval in the same message
-
-Formatting:
-- File links in repo docs/checklists: relative paths, `/`, spaces as `%20`.
-- In assistant UI responses, use the link format required by the execution environment; include relative path text when possible.
-- External links in assistant UI responses: use markdown links with a clear label, not bare URLs.
-
-## Rules Maintenance (MUST)
-- Changes to agent rule files: prefer compressed, machine-readable edits.
-- Keep updates minimal and non-duplicative: merge overlapping points, remove redundancy, preserve intent.
-
-## Memory Log (MUST)
-
-TRIGGER_READ: session_start → call `read_graph` OR `search_nodes(<topic>)`
-TRIGGER_WRITE: significant_change_complete → call `create_entities` OR `add_observations`
-KEY_FORMAT: `<project-prefix>/<entityType-short>/<name>`; prefix defined in project AGENTS.md
-TOPIC_ISOLATION: one entityType per entity; never mix topics in one entity
-ALLOWED_TYPES: architecture | mcp_tool | decision | bug | config | task_log
diff --git a/AGENT.interaction.md b/AGENT.interaction.md
deleted file mode 100644
index 0c8be20..0000000
--- a/AGENT.interaction.md
+++ /dev/null
@@ -1,226 +0,0 @@
-# AGENT.interaction
-
-## ACTIVATION
-
-SCOPE: any agent-initiated user-facing output requiring a choice, approval, confirmation, clarification, branch selection, or action selection.
-
-TRIGGER: automatic — no explicit activation required.
-
-SCOPE EXCLUSIONS: internal tool calls, system/developer-level constraints.
-
-OVERRIDE: these rules override normal conversational behavior within scope. Safety and ethics constraints take absolute precedence.
-
----
-
-## OUTPUT CONTRACT
-
-### Format
-
-```
-<INSTRUCTION>
-
-1) <OPTION>
-2) <OPTION>
-3) <OPTION>
-```
-
-### Constraints
-
-- First character of response = first character of instruction text
-- Instruction: single sentence, ≤ 15 words
-- Options: ≤ 7 per question, ≤ 5 words each
-- Blank line between instruction and options
-- No text before instruction
-- No text after last option
-- No labels: "Answer:", "Choose:", "Single choice"
-- No explanations, comments, status messages, summaries
-- No soft offers (see SOFT OFFERS)
-- No "Other" option (see OPTIONS AUTHORITY)
-
-### Valid input
-
-```
-1          → single select
-1,3        → multi-select (comma-separated, no spaces)
-abort      → emit FLOW_CANCEL
-cancel     → emit FLOW_CANCEL
-<freetext> → pass to system as-is; do not interpret
-```
-
-### Invalid input — reject and repeat
-
-```
-"option 2"
-"I choose 1"
-"probably 3"
-"1, 3"     (spaces in multi-select)
-""         (empty)
-```
-
----
-
-## FLOW
-
-### State machine
-
-```
-step_1 → step_2 → ... → CONFIRMATION
-           ↑_________ RESTART ________|
-```
-
-- One question per step
-- Do not proceed until current step receives valid input
-- Do not skip steps
-- Do not persist state between sessions — state is injected by system per request
-- Do not reorder steps
-
-### Step completion
-
-Step is complete when: input is valid AND value is parsed.
-Otherwise: stay on same step.
-
-### Confirmation (mandatory final step)
-
-```
-Summary:
-- <field>: <value>
-- <field>: <value>
-
-1) Confirm
-2) Restart
-3) Edit specific step
-```
-
-### Compact mode
-
-Input format: `key=N key=N ...` where `key` is the step identifier (e.g. `branch=2 env=1`).
-
-- N must be an exact option number
-- Valid fields: apply, skip their steps
-- Invalid N for a field: reject that field, ask its step normally
-- All fields valid: skip to CONFIRMATION
-
----
-
-## EXECUTION SILENCE
-
-After receiving valid input, the model MUST execute silently (see also: OUTPUT PURITY).
-
-PROHIBITED after valid input:
-- announcing the plan before acting ("Перехожу на ветку X, затем...")
-- narrating steps in progress
-- summarizing what was done in free text
-- any output between receiving input and the next question block or result
-
-The next user-visible output after valid input MUST be either:
-- the result of the action (tool output, file content, etc.), OR
-- the next question block in the flow
-
-```
-✗ "Перехожу на релизную ветку feat/release-v0.2.9, затем подготовлю версию 0.2.9..."
-✓ [executes silently] → next question block or action result
-```
-
----
-
-## VALIDATION
-
-### Strictness
-
-- Exact match only
-- No partial matches
-- No intent interpretation
-- No auto-correction
-- No implicit defaults
-
-### Retry sequence
-
-```
-attempt 1: show error + repeat question
-attempt 2: show error + emphasize format (1 / 2 / 1,3)
-attempt 3: minimal hint
-attempt 4: emit STEP_ABORT
-```
-
-Never relax rules across retries.
-
----
-
-## ABORT & ESCALATION
-
-| Condition | Emit |
-|---|---|
-| 4+ consecutive invalid inputs | `STEP_ABORT` |
-| input = `abort` or `cancel` | `FLOW_CANCEL` |
-
-On abort: emit event code only. No additional text.
-
----
-
-## SOFT OFFERS
-
-PROHIBITED. Any phrase offering an action without presenting a choice block is a format violation.
-
-```
-✗ "Если хочешь, могу выполнить коммит"
-✗ "Let me know if you want to proceed"
-✗ "Могу сделать push — скажи, если нужно"
-```
-
-When an action is available — even one — present a choice block:
-
-```
-Выполнить коммит и push?
-
-1) Да
-2) Нет
-```
-
----
-
-## OPTIONS AUTHORITY
-
-- Option list MUST be exhaustive
-- "Other" is PROHIBITED
-- Free-text is always available to the user implicitly — never offer it as an option
-- An incomplete option list is a design error; resolve by expanding options or splitting into sub-steps
-
----
-
-## OUTPUT PURITY
-
-Extends EXECUTION SILENCE to all responses within scope, not only post-input.
-
-PROHIBITED in any response within scope:
-- Reasoning, planning, analysis
-- Status updates, progress logs
-- Action explanations
-
-Internal reasoning MUST NOT appear in output.
-
----
-
-## DETERMINISM
-
-- Same input → same output
-- Option order: fixed, never shuffled
-- Option text: never rephrased between retries or sessions
-
----
-
-## VIOLATION HANDLING
-
-Non-compliant output: system discards and regenerates.
-Model is not notified of regeneration.
-
----
-
-## PRIORITY
-
-```
-1. Safety & ethics     — absolute, cannot be overridden
-2. Determinism
-3. Structure
-4. Format compliance
-5. User convenience
-```