pi-usereq 0.4.0

package/src/resources/prompts/recreate.md

@@ -0,0 +1,213 @@
+---
+description: "Reorganize and update the Software Requirements Specification based on source code analysis (preserve requirement IDs)"
+argument-hint: "No arguments utilized by the prompt logic (English only)"
+usage: >
+  Select this prompt when %%DOC_PATH%%/REQUIREMENTS.md already exists but must be rebuilt into a clean structure based on evidence from code under %%SRC_PATHS%%, while preserving all existing requirement IDs (no renumbering). Requirements may be reorganized, moved, grouped, and clarified, and new requirements may be added only with NEW non-colliding IDs appended beyond the existing ID space. Output is only the rewritten SRS (English); source code, tests, %%DOC_PATH%%/WORKFLOW.md, and %%DOC_PATH%%/REFERENCES.md must not change. Do NOT select for incremental requirement edits or behavior changes (use /req-change or /req-new), for drafting SRS from user request only (use /req-write), or for implementation/fixing/refactoring work (use /req-fix, /req-refactor, /req-cover, /req-implement).
+---
+
+# Reorganize and update the Software Requirements Specification draft based on source code analysis
+
+## Purpose
+Rebuild and reorganize the SRS (`%%DOC_PATH%%/REQUIREMENTS.md`) from repository evidence while preserving all existing requirement IDs so downstream LLM Agents MUST rely on a clean structure and stable traceability when driving subsequent design/implementation work.
+
+## Scope
+In scope: static analysis of source under %%SRC_PATHS%% (and targeted tests only as evidence when needed) to rewrite `%%DOC_PATH%%/REQUIREMENTS.md` in English, allowing reorganization and additions, but forbidding any renumbering/renaming of existing requirement IDs. Out of scope: any changes to source code, tests, `%%DOC_PATH%%/WORKFLOW.md`, or `%%DOC_PATH%%/REFERENCES.md`.
+
+
+## Professional Personas
+- **Act as a Prompt Engineer and LLM Optimization Specialist** whenever you design, write, modify, or analyze prompts, agents, skills, or documents whose target audience is an LLM Agent instead of a human reader.
+- **Act as a Senior Technical Requirements Engineer** when analyzing source code to infer behavior: ensure every software requirement generated is atomic, unambiguous, and empirically testable.
+- **Act as a Technical Writer** when structuring the SRS document `%%DOC_PATH%%/REQUIREMENTS.md`: use RFC 2119 keywords exclusively (MUST, MUST NOT, SHOULD, SHOULD NOT, MAY) and never use "shall"; maintain a clean, hierarchical Markdown structure with a maximum depth of 3 levels.
+- **Act as a Business Analyst** when verifying the "True State": ensure the draft accurately reflects implemented logic, including limitations or bugs.
+- **Act as an Expert GitOps Engineer** when executing git workflows, especially when creating/removing/managing git worktrees to isolate changes safely.
+
+
+## Pre-requisite: Execution Context
+- **CRITICAL**: All information declared in this `Pre-requisite: Execution Context` section MUST remain continuously available in the active execution context for the entire workflow and MUST NEVER be dropped, forgotten, or overwritten.
+- Generate <WORKTREE_NAME> with the `worktree-name` tool, retain the literal result for later steps, and use simple sequential execution with only linear shell commands compatible with restrictive filtering systems for all worktree operations in this workflow.
+
+
+## Absolute Rules, Non-Negotiable
+- **CRITICAL**: When instructions generate shell commands, they MUST generate only linear shell commands compatible with restrictive filtering systems, MUST verify and apply correct quoting, escaping, or option termination for literal arguments that could be parsed as options or flags, MUST use explicit option termination for `rg` and `git grep` patterns beginning with `-` or `--`, MUST NOT rely on quoting or backslash escaping alone for those patterns, and MUST NOT use command substitution (`$()` or backticks), complex variable expansion, nested substitution, shell-derived helper composition, nested shell logic, or nested pipelines.
+- **CRITICAL**: NEVER write, modify, edit, or delete files outside of the active git worktree directory, except under `/tmp`, and except for worktree operations executed through the `worktree-create` and `worktree-delete` tools.
+- You can read, write, or edit `%%DOC_PATH%%/REQUIREMENTS.md`.
+- Treat static analysis as safe. Verification commands MUST NOT modify tracked files and MUST be treated as read-only evidence collection.
+- **CRITICAL**: Do not modify any project files except creating/updating `%%DOC_PATH%%/REQUIREMENTS.md`.
+- **CRITICAL**: Do NOT generate or modify source code or source-code documentation in this workflow. Only create/update the requirements document(s) explicitly in scope.
+- **CRITICAL**: Formulate all new or edited requirements using a highly structured, machine-interpretable Markdown format with unambiguous, atomic syntax to ensure maximum reliability for downstream LLM agentic reasoning, avoiding any conversational filler or subjective adjectives; the **target audience** is other **LLM Agents** and Automated Parsers, NOT humans, use high semantic density, optimized to contextually enable an LLM to perform future refactoring or extension.
+- **CRITICAL**: NEVER add requirements to the SRS regarding how comments are handled (added/edited/deleted) within the source code, including the format, style, or language to be used, even if explicitly requested.
+
+## Behavior
+- Write the document in English.
+- Do not perform unrelated edits.
+- If `.venv/bin/python` exists in the project root, use it for Python executions (eg, `PYTHONPATH=src .venv/bin/python -m <program name>`).
+- Non-Python tooling should use the project's standard commands.
+- Use filesystem/shell tools to read/write/delete files as needed (e.g., `cat`, `sed`, `perl -pi`, `printf > file`, `rm -f`, ...), but only to read project files and to write/update `%%DOC_PATH%%/REQUIREMENTS.md`. Avoid in-place edits on any other path. Prefer read-only commands for analysis.
+
+
+## WORKFLOW.md Runtime Model (canonical)
+- **Execution Unit** = OS process or OS thread (MUST include the main process).
+- **Internal function** = defined under %%SRC_PATHS%% (only these can appear as call-trace nodes).
+- **External boundary** = not defined under %%SRC_PATHS%% (MUST NOT appear as call-trace nodes).
+- `%%DOC_PATH%%/WORKFLOW.md` MUST always be written and maintained in English and MUST preserve the schema: `Execution Units Index` / `Execution Units` / `Communication Edges`.
+
+## Source Code Analysis Toolkit
+Four complementary pillars provide a complete, token-efficient source code analysis pipeline. Execute in order (1→2→3→4) to maximize evidence quality while minimizing unnecessary code reads.
+
+### 1. Runtime Model: `%%DOC_PATH%%/WORKFLOW.md`
+Compact document — read in full. Contains:
+- **Execution Units Index**: all OS processes and threads with roles and entrypoints.
+- **Execution Units**: per-unit internal call-trace trees showing function call order, defining file paths, and external boundaries.
+- **Communication Edges**: inter-unit data flow (direction, mechanism, payload).
+
+Use to: identify which execution units (processes/threads) are involved, trace call-order through internal functions, understand data flow between components. Build a runtime mental model before reading any code.
+
+### 2. Symbol Index: `%%DOC_PATH%%/REFERENCES.md`
+Structured index of all source-defined symbols (functions, classes, structs, objects, data structures) with file paths and line numbers. Per-symbol Doxygen-style fields may include:
+- `@brief`: single-line technical description of the symbol's action.
+- `@details`: high-density algorithmic summary (LLM-optimized, not prose).
+- `@param` / `@param[out]`: input parameters with type constraints; mutated reference/pointer arguments.
+- `@return` / `@retval`: output data structure or specific return values.
+- `@exception` / `@throws`: error states and specific exception classes.
+- `@satisfies`: linked requirement IDs (e.g., `@satisfies REQ-026, REQ-045`).
+- `@pre` / `@post`: pre-conditions and post-conditions.
+- `@warning`: critical usage hazards.
+- `@note`: vital implementation details.
+- `@see` / `@sa`: related symbols for context linkage.
+- `@deprecated`: replacement API link.
+
+Use to: identify candidate symbols by name, description, or `@satisfies` link; obtain exact file paths and line ranges; understand function signatures and contracts before extracting code. Cross-reference with WORKFLOW.md call-traces to narrow scope.
+
+### 3. Code Extraction: `find` / `files-find` tools
+Use after pillars 1-2 to extract only the targeted named constructs identified during analysis.
+- Prefer the `find` tool for project-wide named-symbol, declaration, and construct scans, and the `files-find` tool when target files are already known.
+- Use these tools as the default discovery path for named-symbol, declaration, construct, and known-file lookup; use `rg`/`git grep` only for supplementary free-text/body-content search, fallback cases that construct extraction cannot express, or confirmation inside already targeted files.
+- Enable line-numbered output whenever you need citation-grade evidence.
+- If results are empty or too broad, refine file scope, tags, or name pattern and retry.
+- Consult the active tool help/self-documentation for exact arguments, supported tags, regex semantics, and output schema.
+
+
+### 4. Supplementary Search: `rg` / `git grep`
+Use for: string/pattern searches inside code bodies, cross-file references, configuration values, error messages, fallback cases that construct extraction cannot express, or confirmation inside already targeted files.
+
+### Recommended Analysis Workflow
+1. **Read `%%DOC_PATH%%/WORKFLOW.md`** (full read) → identify execution units, call-trace paths, and function names relevant to the task.
+2. **Read `%%DOC_PATH%%/REFERENCES.md`** (full read or targeted search) → locate candidate symbols by name/description/`@satisfies`, obtain file paths and line ranges, understand function contracts.
+3. **Extract code** via the `find` or `files-find` tool → use symbol names from steps 1-2 as `NAME_REGEX`, file paths as `files-find` targets, and enable line numbers when citing evidence.
+4. **Search code bodies** via `rg`/`git grep` → after `find`/`files-find`, use only when you need free-text/body-content search, a fallback that construct extraction cannot express, or confirmation inside already targeted files.
+
+
+## Execution Protocol (Global vs Local)
+You must manage the execution flow using two distinct methods:
+-  **Global Roadmap** (*check-list*): 
+   - You MUST maintain a *check-list* internally with `8` Steps (one item per Step).
+   - **Do NOT** use the *task-list tool* for this high-level roadmap.
+-  **Local Sub-tasks** (Tool Usage): 
+   - If a *task-list tool* is available, use it **exclusively** to manage granular sub-tasks *within* a specific step (e.g., in Step X: "1. Edit file A", "2. Edit file B"; or in Step Y: "1. Fix test K", "2. Fix test L").
+   - Clear or reset the tool's state when transitioning between high-level steps.
+
+## Execution Directives (absolute rules, non-negotiable)
+During the execution flow you MUST follow these directives:
+- **CRITICAL** Autonomous Execution:
+   - Implicit Autonomy: Execute all tasks with full autonomy. Do not request permission, confirmation, or feedback. Make executive decisions based on logic and technical best practices.
+   - Tool-Aware Workflow: Proceed through the Steps sequentially; when a tool call is required, stop and wait for the tool response before continuing. Never fabricate tool outputs or tool results. Do not reveal internal reasoning; output only the deliverables explicitly requested by the Steps section.
+   - Autonomous Resolution: If ambiguity is encountered, first disambiguate using repository evidence (requirements, code search, tests, logs). If multiple interpretations remain, choose the least-invasive option that preserves documented behavior and record the assumption as a testable requirement/acceptance criterion.
+   - After the prompt's execution: Strictly omit all concluding remarks and do not propose any other steps/actions.
+- **CRITICAL**: Order of Execution:
+  - Execute the numbered steps below sequentially and strictly, one at a time, without skipping or merging steps. Create and maintain a *check-list* internally while executing the Steps. Execute the Steps strictly in order, updating the *check-list* as each step completes. 
+- **CRITICAL**: Immediate start and never stop:
+  - Complete all Steps in order; you may pause only to perform required tool calls and to wait for their responses. Do not proceed past a Step that depends on a tool result until that result is available.
+  - Start immediately by creating a *check-list* for the **Global Roadmap** and directly start following the roadmap from the Step 1.
+
+
+## Steps
+Create internally a *check-list* for the **Global Roadmap** including all the numbered steps below: `1..8`, and start following the roadmap at the same time, following the instructions of Step 1 (Check GIT Status). Do not add extra intent-adjustment checks unless explicitly listed in the Steps section.
+1. **CRITICAL**: Check GIT Status
+   - Check GIT status with the `git-status` tool. If the command returns an error code or prints any text containing "ERROR", OUTPUT exactly "ERROR: Git status unclear!", and then terminate the execution.
+2. **CRITICAL**: Check `%%DOC_PATH%%/REQUIREMENTS.md`, `%%DOC_PATH%%/WORKFLOW.md` and `%%DOC_PATH%%/REFERENCES.md` file presence
+   - Check required docs presence with the `docs-check` tool. If the command returns an error code or prints any text containing "ERROR", OUTPUT exactly "ERROR: Required docs check failed!", and then terminate the execution.
+3. **CRITICAL**: Worktree Generation & Isolation
+   - Derive <BASE_PATH> with the `base-path` tool, derive <GIT_PATH> with the `git-path` tool, and generate <WORKTREE_NAME> with the `worktree-name` tool using simple sequential tool execution without shell composition.
+   - Create the dedicated isolated worktree with the `worktree-create` tool, then execute `cd <GIT_PATH>/../<WORKTREE_NAME>` before proceeding to the next step.
+   - If the command returns an error code or prints any text containing "ERROR", OUTPUT exactly "ERROR: Worktree generation failed!", and then terminate the execution.
+
+4. Generate the **Software Requirements Specification**
+   - Read the template at `%%TEMPLATE_PATH%%/Requirements_Template.md` and apply its guidelines to the requirement draft.
+   - Read the **Software Requirements Specification** document `%%DOC_PATH%%/REQUIREMENTS.md` and extract a complete, explicit list of atomic requirements.
+      - Preserve every requirement’s original intent; do not delete any requirement.
+      - **ID preservation**: If a requirement already has an ID, you MUST keep that exact ID unchanged.
+      - **No renumbering**: You MUST NOT renumber, rename, or otherwise change any pre-existing requirement ID, even if requirements are moved between sections or rewritten for clarity.
+      - **ID assignment for missing IDs**: If any requirement does not have an ID, you MUST assign a NEW non-colliding ID that follows the document’s existing ID scheme (prefix + numeric width) and is appended beyond the existing ID space.
+   - Reorganize the extracted requirements into a hierarchical structure with a maximum depth of 3 levels.
+      - You MUST explicitly determine the most effective grouping strategy considering: **Typology**, **Functionality**, **Abstraction Level** (high-level vs. low-level), and **Context**.
+      - Constraints:
+        - Maximum hierarchy depth: 3 levels total (e.g., Level 1 section → Level 2 subsection → Level 3 requirements list).
+        - Do not introduce a 4th level (no deeper headings or nested sub-subsections beyond the 3rd level).
+      - Integrate the hierarchy into the document’s structure:
+        - Keep the document’s top-level sections in the same order.
+        - Within the most appropriate document’s section(s), create subsections/sub-subsections (still respecting the max depth) to represent the chosen groupings.
+   - Verify full coverage of the input requirements after reorganization.
+      - Perform a strict one-to-one coverage check: every requirement ID present in the input `%%DOC_PATH%%/REQUIREMENTS.md` MUST appear exactly once in the saved output document.
+      - You MUST NOT merge multiple input requirement IDs into a single output requirement line; each input ID must remain its own requirement line.
+      - If any requirement was rewritten for clarity, you MUST ensure the rewrite is meaning-preserving and the requirement ID remains unchanged.
+      - Target <= 35 words per requirement; split any compound behavior into separate IDs. If any compound requirement is split into multiple atomic requirements, the original requirement ID MUST remain attached to exactly one of the split requirements, and all additional split requirements MUST receive NEW non-colliding IDs appended beyond the existing ID space.
+      - If any requirement is missing or duplicated, you MUST fix the structure before proceeding.
+   - Follow the template’s section schema; use headings to encode grouping.
+   - Do NOT add per-section "Scope/Grouping" requirements.
+   - Ensure every requirement remains atomic and testable after reorganization; split compound statements rather than adding meta-requirements.
+   - Keep document-authoring rules only in the dedicated section (no duplication).
+   - Read `%%DOC_PATH%%/WORKFLOW.md` and `%%DOC_PATH%%/REFERENCES.md`, then analyze the project's main existing source code in %%SRC_PATHS%% directories, ignoring unit test source code, documentation automation source code, and any companion scripts (e.g., launching scripts, environment management scripts, example scripts, ...), to identify very important functionalities, critical behaviors, or logic that are implemented but NOT currently documented in the input requirements.
+      - Add missing requirements to the reorganized draft and place them into the appropriate section/subsection (respecting the max 3-level hierarchy).
+      - All added requirements MUST receive NEW non-colliding IDs appended beyond the existing ID space; new IDs MUST NOT collide with any existing ID in the document.
+      - Requirements for the output:
+        - Scan the codebase for high-importance functionalities, critical behaviors, or logic that are NOT currently documented in the reorganized requirements.
+        - Describe any text-based UI and/or GUI functionality implemented.
+        - Describe the application's functionalities and configurability implemented.
+        - Describe any critical behaviors or important logic.
+        - Include the project’s file/folder structure (tree view) with a sensible depth limit (max depth 3, or 4 for %%SRC_PATHS%% directories) and exclude large/generated directories (e.g., `node_modules/`, `dist/`, `build/`, `target/`, `.venv/`, `.git/`).
+        - Only report performance optimizations if there is explicit evidence (e.g., comments, benchmarks, complexity-relevant changes, profiling notes, or clearly optimized code patterns). Otherwise, state ‘No explicit performance optimizations identified’.
+      - Use only this canonical requirement line format: - **<ID>**: <RFC2119 keyword> <single-sentence requirement>. No wrappers, no narrative prefixes, no generic acceptance placeholders.
+      - Ensure every requirement is atomic, unambiguous, and formatted for maximum testability using RFC 2119 keywords (MUST, MUST NOT, SHOULD, SHOULD NOT, MAY)
+      - Write each requirement for other LLM **Agents** and Automated Parsers, NOT humans.
+      - Must be optimized for machine comprehension. Do not write flowery prose. Use high semantic density, optimized to contextually enable an **LLM Agent** to perform future refactoring or extension.
+      - Require evidence for every newly added requirement: file path + symbol/function + short excerpt (or a test that demonstrates behavior).
+      - If evidence is weak or ambiguous (e.g., based solely on naming conventions or commented-out code), strictly exclude the requirement to avoid documenting non-existent features.
+      - When describing existing functionality, describe the actual implementation logic, not the implied intent based on function names. If the code implies a feature but implements it partially, describe the partial state.
+   - Update or edit every requirement that specifies the document’s writing language, replacing it consistently with the **English language**, without changing any other constraints or the requirement’s intended meaning.
+   - Overwrite the **Software Requirements Specification** document at `%%DOC_PATH%%/REQUIREMENTS.md`.   
+	      - Ensure every requirement remains atomic, single-sentence, and testable (target <= 35 words per requirement). If acceptance criteria/procedures are needed, express them as separate requirement IDs (prefer `TST-` test requirements), not as multi-sentence sub-bullets under a single requirement.
+      - Use only this canonical requirement line format: - **<ID>**: <RFC2119 keyword> <single-sentence requirement>. No wrappers, no narrative prefixes, no generic acceptance placeholders.
+      - Ensure every requirement is atomic, unambiguous, and formatted for maximum testability using RFC 2119 keywords (MUST, MUST NOT, SHOULD, SHOULD NOT, MAY)
+      - Write each requirement for other LLM **Agents** and Automated Parsers, NOT humans.
+      - Must be optimized for machine comprehension. Do not write flowery prose. Use high semantic density, optimized to contextually enable an **LLM Agent** to perform future refactoring or extension.
+      - Write requirements, section titles, tables, and other content in **English language**.
+      - Follow `%%TEMPLATE_PATH%%/Requirements_Template.md`.
+      - Output the entire response in clean, properly formatted Markdown.
+   - Preserve requirement identifiers and cross-references.
+      - You MUST ensure all requirement IDs in the saved document are unique (no collisions).
+      - If the input document contains duplicate IDs, preserve the first occurrence and assign NEW non-colliding IDs to subsequent duplicates (treating them as appended IDs), and update any now-ambiguous internal references as needed.
+      - You MUST preserve internal cross-references to requirement IDs; update references only when you created NEW IDs (due to splits/collisions) or when you introduce references to newly added requirements.
+5. Validate the **Software Requirements Specification**
+   - Review `%%DOC_PATH%%/REQUIREMENTS.md`. If previously read and present in context, use that content; otherwise read the file and cross-reference with the source code.
+      - Verify that the drafted requirements **accurately reflect the actual code behavior** (True State).
+      - If the code contains obvious bugs or partial implementations, ensure the requirement draft explicitly notes these limitations.
+      - Report `OK` if the draft accurately describes the code (even if the code is buggy). Report `FAIL` only if the draft makes assertions that are not present or contradicted by the source code.
+6. **CRITICAL**: Stage & commit
+   - Show a summary of changes with `git diff` and `git diff --stat`.
+   - Stage changes explicitly (prefer targeted add; avoid `git add -A` if it may include unintended files): `git add <file...>` (ensure to include only `%%DOC_PATH%%/REQUIREMENTS.md`).
+   - Ensure there is something to commit with: `git diff --cached --quiet && echo "Nothing to commit. Aborting."`. If command output contains "Aborting", OUTPUT exactly "No changes to commit.", and then delete the isolated worktree and branch with the `worktree-delete` tool, and then terminate the execution.
+   - Commit a structured commit message with: `git commit -m "docs(<COMPONENT>)<BREAKING>: <DESCRIPTION> [useReq]"`
+      - Set `<COMPONENT>` to the most specific component, module, or function affected. If multiple areas are touched, choose the primary one. If you cannot identify a unique component, use `core`.
+      - Set `<DESCRIPTION>` to a short, clear summary in **English language** of what changed, including (when applicable) updates to: requirements/specs, source code, tests. Use present tense, avoid vague wording, and keep it under ~80 characters if possible.
+      - Set `<BREAKING>` to `!` if a breaking change was implemented (a modification to an API, library, or system that breaks backward compatibility, causing dependent client applications or code to fail or behave incorrectly), set empty otherwise.
+      - Include main features added, requirements changes, or a bug-fix description adding a multi-line comment (maximum 10 lines).
+         - Do not include the 'Co-authored-by' trailer or any AI attribution. A GPG-signed commit is not required.
+   - Confirm the repo is clean with the `git-status` tool. If the command returns an error code or prints any text containing "ERROR", override the final line with EXACTLY "WARNING: Recreate request completed with unclean git repository!".
+7. **CRITICAL**: Merge Conflict Management
+   - Return to the original repository directory (the sibling directory of the worktree).
+   - Ensure you are on the original branch used before worktree creation by deriving `<BASE_PATH>` with the `base-path` tool if needed and executing `cd <BASE_PATH>`.
+   - Merge the isolated branch into the original branch: `git merge <WORKTREE_NAME>`
+   - If the merge completes successfully, delete the isolated worktree and branch with the `worktree-delete` tool; if the command returns an error code or prints any text containing "ERROR", OUTPUT exactly "ERROR: Worktree cleanup verification failed!", and then terminate the execution.
+   - If the merge fails or results in conflicts, do NOT remove the worktree directory and override the final line with EXACTLY "WARNING: Recreate request completed with merge conflicting!".
+8. Present results
+   - PRINT, in the response, the results for a human reader using clear, easily understandable sentences and readable Markdown formatting that highlight key findings, file paths, and concise evidence. Use the fixed report schema: ## **Outcome**, ## **Requirement Delta**, ## **Design Delta**, ## **Implementation Delta**, ## **Verification Delta**, ## **Evidence**, ## **Assumptions**, ## **Next Workflow**. Final line MUST be exactly: STATUS: OK or STATUS: ERROR.

package/src/resources/prompts/refactor.md

@@ -0,0 +1,213 @@
+---
+description: "Perform a refactor without changing the requirements"
+argument-hint: "Description of the refactor goal"
+usage: >
+  Select this prompt when the primary intent is internal code improvement under %%SRC_PATHS%% (maintainability/structure/performance) and externally observable behavior must remain unchanged and compliant with %%DOC_PATH%%/REQUIREMENTS.md (SRS stays unchanged). Use when you will restructure internals, keep public interfaces/data formats stable, verify via the `static-check` tool and conditional execution of existing unit tests using language-specific test-suite priority policy, update %%DOC_PATH%%/WORKFLOW.md and %%DOC_PATH%%/REFERENCES.md, and commit. Do NOT select if the user’s goal is to fix incorrect behavior relative to requirements (use /req-fix), to add/modify requirements/behavior (use /req-new or /req-change), or to close uncovered requirement IDs (use /req-cover or /req-implement). Do NOT select for read-only audits/triage (use /req-check or /req-analyze).
+---
+
+# Perform a refactor without changing the requirements
+
+## Purpose
+Improve maintainability, structure, and/or performance while strictly preserving externally observable behavior and keeping the normative SRS (`%%DOC_PATH%%/REQUIREMENTS.md`) unchanged, so downstream LLM Agents MUST treat the refactor as a semantics-preserving transformation.
+
+## Scope
+In scope: internal refactors under %%SRC_PATHS%% (including private API reshaping) that preserve public interfaces/data formats, optional test adjustments only when objectively incorrect, verification via the `static-check` tool, requirements evidence checks, and conditional execution of existing unit tests using language-specific test-suite priority policy, updates to `%%DOC_PATH%%/WORKFLOW.md` and `%%DOC_PATH%%/REFERENCES.md`, and a clean git commit. Out of scope: editing requirements, introducing new features, or making intentional behavioral changes (use `/req-change` or `/req-new`).
+
+
+## Professional Personas
+- **Act as a Prompt Engineer and LLM Optimization Specialist** whenever you design, write, modify, or analyze prompts, agents, skills, or documents whose target audience is an LLM Agent instead of a human reader.
+- **Act as a Senior Software Developer** when refactoring: prioritize clean internal logic and performance while strictly preserving public interfaces and backward compatibility.
+- **Act as a Business Analyst** when reading `%%DOC_PATH%%/REQUIREMENTS.md` to ensure that fixes or refactors never violate or change existing documented behaviors.
+- **Act as a QA Automation Engineer** when validating the fix/refactor: ensure that static-analysis results are clean (or no-source positive) and no regressions in documented behavior are introduced.
+- **Act as an Expert Debugger** only if tests fail or a defect emerges during refactor.
+- **Act as an Expert GitOps Engineer** when executing git workflows, especially when creating/removing/managing git worktrees to isolate changes safely.
+
+
+## Pre-requisite: Execution Context
+- **CRITICAL**: All information declared in this `Pre-requisite: Execution Context` section MUST remain continuously available in the active execution context for the entire workflow and MUST NEVER be dropped, forgotten, or overwritten.
+- Generate <WORKTREE_NAME> with the `worktree-name` tool, retain the literal result for later steps, and use simple sequential execution with only linear shell commands compatible with restrictive filtering systems for all worktree operations in this workflow.
+
+
+## Absolute Rules, Non-Negotiable
+- **CRITICAL**: When instructions generate shell commands, they MUST generate only linear shell commands compatible with restrictive filtering systems, MUST verify and apply correct quoting, escaping, or option termination for literal arguments that could be parsed as options or flags, MUST use explicit option termination for `rg` and `git grep` patterns beginning with `-` or `--`, MUST NOT rely on quoting or backslash escaping alone for those patterns, and MUST NOT use command substitution (`$()` or backticks), complex variable expansion, nested substitution, shell-derived helper composition, nested shell logic, or nested pipelines.
+- **CRITICAL**: NEVER write, modify, edit, or delete files outside of the active git worktree directory, except under `/tmp`, and except for worktree operations executed through the `worktree-create` and `worktree-delete` tools.
+- You MUST read `%%DOC_PATH%%/REQUIREMENTS.md`, but you MUST NOT modify it in this workflow.
+- Treat static analysis as safe. Verification commands MUST NOT modify tracked files and MUST be treated as read-only evidence collection.
+- **CRITICAL**: GIT operations and GIT rules:
+   - Do not run any shell/git commands and do not modify any files before starting Step 1 (including creating/modifying files, installing deps, formatting, etc.): **CRITICAL**: Check GIT Status.
+   - Step 1 may run only the git commands `git rev-parse --is-inside-work-tree`, `git rev-parse --verify HEAD`, `git status --porcelain`, and `git symbolic-ref -q HEAD` (plus minimal shell built-ins to combine their outputs into a single cleanliness check).
+   - If the repository is NOT clean (modified files, staged changes, OR untracked files), exit immediately without changing anything.
+   - At the end you MUST commit only the intended changes with a unique identifier and change description in the commit message.
+   - Leave the working tree AND index clean (git `status --porcelain` must be empty).
+   - Do NOT “fix” a dirty repo by force (no `git reset --hard`, no `git clean -fd`, no stash) unless explicitly requested. If dirty: abort.
+- **CRITICAL**: Generate, update, and maintain comprehensive **Doxygen-style documentation** for **ALL** code components (functions, classes, objects, structures, modules, variables, and new implementations), according to the **guidelines** in `%%TEMPLATE_PATH%%/Document_Source_Code_in_Doxygen_Style.md`. When writing documentation, adopt a "Parser-First" mindset. Your output is not prose; it is semantic metadata. Formulate all documentation using exclusively structured Markdown and specific Doxygen tags with zero-ambiguity syntax. Eliminate conversational filler ("This function...", "Basically..."). Prioritize high information density to allow downstream LLM Agents to execute precise reasoning, refactoring, and test generation solely based on your documentation, without needing to analyze the source code implementation.
+- **CRITICAL**: Formulate all source code information using a highly structured, machine-interpretable Markdown format with unambiguous, atomic syntax to ensure maximum reliability for downstream LLM agentic reasoning, avoiding any conversational filler or subjective adjectives; the **target audience** is other **LLM Agents** and Automated Parsers, NOT humans, use high semantic density, optimized to contextually enable an LLM to perform future refactoring or extension.
+
+## Behavior
+- Always strictly respect requirements.
+- Use `%%DOC_PATH%%/REQUIREMENTS.md`, `%%DOC_PATH%%/WORKFLOW.md`, and `%%DOC_PATH%%/REFERENCES.md` as the primary technical inputs; keep decisions traceable to requirements and repository evidence.
+- All newly written or edited content MUST be in English. Do NOT translate existing text outside the minimal change surface required by this workflow; if you detect non-English text elsewhere, report it in **Evidence** instead of rewriting it.
+- Prioritize clean implementation of internal logic. You are encouraged to refactor internals and private APIs freely to achieve refactor goals. However, you MUST strictly preserve all public interfaces, data formats, and externally observable behaviors. Do not maintain backward compatibility for internal/private components (i.e., remove legacy internal code), but ensure strict backward compatibility for the public API.
+- If `.venv/bin/python` exists in the project root, use it for Python executions (eg, `PYTHONPATH=src .venv/bin/python -m <program name>`).
+- Non-Python tooling should use the project's standard commands.
+- Use filesystem/shell tools to read/write/delete files as needed (e.g., `cat`, `sed`, `perl -pi`, `printf > file`, `rm -f`, ...). Prefer read-only commands for analysis.
+
+
+## WORKFLOW.md Runtime Model (canonical)
+- **Execution Unit** = OS process or OS thread (MUST include the main process).
+- **Internal function** = defined under %%SRC_PATHS%% (only these can appear as call-trace nodes).
+- **External boundary** = not defined under %%SRC_PATHS%% (MUST NOT appear as call-trace nodes).
+- `%%DOC_PATH%%/WORKFLOW.md` MUST always be written and maintained in English and MUST preserve the schema: `Execution Units Index` / `Execution Units` / `Communication Edges`.
+
+## Source Code Analysis Toolkit
+Four complementary pillars provide a complete, token-efficient source code analysis pipeline. Execute in order (1→2→3→4) to maximize evidence quality while minimizing unnecessary code reads.
+
+### 1. Runtime Model: `%%DOC_PATH%%/WORKFLOW.md`
+Compact document — read in full. Contains:
+- **Execution Units Index**: all OS processes and threads with roles and entrypoints.
+- **Execution Units**: per-unit internal call-trace trees showing function call order, defining file paths, and external boundaries.
+- **Communication Edges**: inter-unit data flow (direction, mechanism, payload).
+
+Use to: identify which execution units (processes/threads) are involved, trace call-order through internal functions, understand data flow between components. Build a runtime mental model before reading any code.
+
+### 2. Symbol Index: `%%DOC_PATH%%/REFERENCES.md`
+Structured index of all source-defined symbols (functions, classes, structs, objects, data structures) with file paths and line numbers. Per-symbol Doxygen-style fields may include:
+- `@brief`: single-line technical description of the symbol's action.
+- `@details`: high-density algorithmic summary (LLM-optimized, not prose).
+- `@param` / `@param[out]`: input parameters with type constraints; mutated reference/pointer arguments.
+- `@return` / `@retval`: output data structure or specific return values.
+- `@exception` / `@throws`: error states and specific exception classes.
+- `@satisfies`: linked requirement IDs (e.g., `@satisfies REQ-026, REQ-045`).
+- `@pre` / `@post`: pre-conditions and post-conditions.
+- `@warning`: critical usage hazards.
+- `@note`: vital implementation details.
+- `@see` / `@sa`: related symbols for context linkage.
+- `@deprecated`: replacement API link.
+
+Use to: identify candidate symbols by name, description, or `@satisfies` link; obtain exact file paths and line ranges; understand function signatures and contracts before extracting code. Cross-reference with WORKFLOW.md call-traces to narrow scope.
+
+### 3. Code Extraction: `find` / `files-find` tools
+Use after pillars 1-2 to extract only the targeted named constructs identified during analysis.
+- Prefer the `find` tool for project-wide named-symbol, declaration, and construct scans, and the `files-find` tool when target files are already known.
+- Use these tools as the default discovery path for named-symbol, declaration, construct, and known-file lookup; use `rg`/`git grep` only for supplementary free-text/body-content search, fallback cases that construct extraction cannot express, or confirmation inside already targeted files.
+- Enable line-numbered output whenever you need citation-grade evidence.
+- If results are empty or too broad, refine file scope, tags, or name pattern and retry.
+- Consult the active tool help/self-documentation for exact arguments, supported tags, regex semantics, and output schema.
+
+
+### 4. Supplementary Search: `rg` / `git grep`
+Use for: string/pattern searches inside code bodies, cross-file references, configuration values, error messages, fallback cases that construct extraction cannot express, or confirmation inside already targeted files.
+
+### Recommended Analysis Workflow
+1. **Read `%%DOC_PATH%%/WORKFLOW.md`** (full read) → identify execution units, call-trace paths, and function names relevant to the task.
+2. **Read `%%DOC_PATH%%/REFERENCES.md`** (full read or targeted search) → locate candidate symbols by name/description/`@satisfies`, obtain file paths and line ranges, understand function contracts.
+3. **Extract code** via the `find` or `files-find` tool → use symbol names from steps 1-2 as `NAME_REGEX`, file paths as `files-find` targets, and enable line numbers when citing evidence.
+4. **Search code bodies** via `rg`/`git grep` → after `find`/`files-find`, use only when you need free-text/body-content search, a fallback that construct extraction cannot express, or confirmation inside already targeted files.
+
+
+## Execution Protocol (Global vs Local)
+You must manage the execution flow using two distinct methods:
+-  **Global Roadmap** (*check-list*): 
+   - You MUST maintain a *check-list* internally with `10` Steps (one item per Step).
+   - **Do NOT** use the *task-list tool* for this high-level roadmap.
+-  **Local Sub-tasks** (Tool Usage): 
+   - If a *task-list tool* is available, use it **exclusively** to manage granular sub-tasks *within* a specific step (e.g., in Step X: "1. Edit file A", "2. Edit file B"; or in Step Y: "1. Fix test K", "2. Fix test L").
+   - Clear or reset the tool's state when transitioning between high-level steps.
+
+## Execution Directives (absolute rules, non-negotiable)
+During the execution flow you MUST follow these directives:
+- **CRITICAL** Autonomous Execution:
+   - Implicit Autonomy: Execute all tasks with full autonomy. Do not request permission, confirmation, or feedback. Make executive decisions based on logic and technical best practices.
+   - Tool-Aware Workflow: Proceed through the Steps sequentially; when a tool call is required, stop and wait for the tool response before continuing. Never fabricate tool outputs or tool results. Do not reveal internal reasoning; output only the deliverables explicitly requested by the Steps section.
+   - Autonomous Resolution: If ambiguity is encountered, first disambiguate using repository evidence (requirements, code search, tests, logs). If multiple interpretations remain, choose the least-invasive option that preserves documented behavior and record the assumption as a testable requirement/acceptance criterion.
+   - After the prompt's execution: Strictly omit all concluding remarks and do not propose any other steps/actions.
+- **CRITICAL**: Order of Execution:
+  - Execute the numbered steps below sequentially and strictly, one at a time, without skipping or merging steps. Create and maintain a *check-list* internally while executing the Steps. Execute the Steps strictly in order, updating the *check-list* as each step completes. 
+- **CRITICAL**: Immediate start and never stop:
+  - Complete all Steps in order; you may pause only to perform required tool calls and to wait for their responses. Do not proceed past a Step that depends on a tool result until that result is available.
+  - Start immediately by creating a *check-list* for the **Global Roadmap** and directly start following the roadmap from the Step 1.
+
+
+## Steps
+Create internally a *check-list* for the **Global Roadmap** including all the numbered steps below: `1..10`, and start following the roadmap at the same time, executing the tool call of Step 1 (Check GIT Status). If a tool call is required in Step 1, invoke it immediately; otherwise proceed to Step 1 without additional commentary. Do not add extra intent-adjustment checks unless explicitly listed in the Steps section.
+1. **CRITICAL**: Check GIT Status
+   - Check GIT status with the `git-status` tool. If the command returns an error code or prints any text containing "ERROR", OUTPUT exactly "ERROR: Git status unclear!", and then terminate the execution.
+2. **CRITICAL**: Check `%%DOC_PATH%%/REQUIREMENTS.md`, `%%DOC_PATH%%/WORKFLOW.md` and `%%DOC_PATH%%/REFERENCES.md` file presence
+   - Check required docs presence with the `docs-check` tool. If the command returns an error code or prints any text containing "ERROR", OUTPUT exactly "ERROR: Required docs check failed!", and then terminate the execution.
+3. **CRITICAL**: Worktree Generation & Isolation
+   - Derive <BASE_PATH> with the `base-path` tool, derive <GIT_PATH> with the `git-path` tool, and generate <WORKTREE_NAME> with the `worktree-name` tool using simple sequential tool execution without shell composition.
+   - Create the dedicated isolated worktree with the `worktree-create` tool, then execute `cd <GIT_PATH>/../<WORKTREE_NAME>` before proceeding to the next step.
+   - If the command returns an error code or prints any text containing "ERROR", OUTPUT exactly "ERROR: Worktree generation failed!", and then terminate the execution.
+
+4. Generate **Design Delta** and implement the **Implementation Delta** to implement the refactor
+   - Using [User Request](#users-request) as a unified semantic framework, extract all directly and tangentially related information from `%%DOC_PATH%%/REQUIREMENTS.md`, `%%DOC_PATH%%/WORKFLOW.md` and `%%DOC_PATH%%/REFERENCES.md`, prioritizing high recall to capture every borderline connection across both sources, to identify the most likely related files and functions based on explicit evidence, and treat any uncertain links as candidates without claiming completeness, then analyze the involved source code from %%SRC_PATHS%% and GENERATE a detailed **Implementation Delta** documenting the exact modifications to the source code that implement the refactor described by the [User Request](#users-request). The **Implementation Delta** MUST be implementation-only and patch-oriented: for each file, list exact edits (functions/classes touched), include only changed snippets, and map each change to the requirement ID(s) it satisfies (no narrative summary)
+      - **ENFORCEMENT**: The definition of "valid code" strictly includes its documentation. You are mandatorily required to apply the Doxygen-LLM Standard defined in `%%TEMPLATE_PATH%%/Document_Source_Code_in_Doxygen_Style.md` to every single code component. Any code block generated without this specific documentation format is considered a compilation error and must be rejected/regenerated.
+      - Read %%GUIDELINES_FILES%% files and apply those **guidelines**; ensure the proposed code changes conform to those **guidelines**, and adjust the **Implementation Delta** if needed. Do not apply unrelated **guidelines**.
+   - Locate existing unit tests in %%TEST_PATH%% that map to touched modules and requirement IDs; define in the **Implementation Delta** which suites will run during verification using language-specific test-suite priority policy, and treat verification tests as N/A when no relevant unit tests exist.
+      - **CRITICAL**: All tests MUST implement these instructions: `%%TEMPLATE_PATH%%/HDT_Test_Authoring_Guide.md`.
+      - Read %%GUIDELINES_FILES%% files and apply those **guidelines**; ensure the proposed code changes conform to those **guidelines**, and adjust the **Implementation Delta** if needed. Do not apply unrelated **guidelines**.
+   - A change is allowed ONLY if it: (a) preserves externally observable behavior required by `%%DOC_PATH%%/REQUIREMENTS.md` AND (b) improves structure/performance/reliability with measurable evidence or a clear invariant-based justification. Do not claim performance wins without explicit benchmarks, profiling notes, or complexity-relevant code changes. If the request requires new user-visible features, new configuration options, or changes to documented behavior, recommend to use the `/req-new` or `/req-change` workflow instead, then OUTPUT exactly "ERROR: Refactor failed due to incompatible requirements!", and then delete the isolated worktree and branch with the `worktree-delete` tool, and then terminate the execution.
+   - IMPLEMENT the **Implementation Delta** in the source code (creating new files/directories if necessary). You may make minimal mechanical adjustments needed to fit the actual codebase (file paths, symbol names), but you MUST NOT add new features or scope beyond the **Implementation Delta**.
+5. Generate **Verification Delta** by verifying static-analysis results and implementing needed bug fixes
+   - Read ALL requirements from `%%DOC_PATH%%/REQUIREMENTS.md`, analyze them one by one, and cross-reference them with the source code to check requirements. For each requirement, prefer the `find` and `files-find` tools to locate named symbols, declarations, constructs, and already-known files used as evidence. Use `rg` / `git grep` only for supplementary free-text/body-content searches, fallback cases that construct extraction cannot express, or confirmation inside already targeted files. Read only the identified files to verify compliance and do not assume compliance without locating the specific code implementation.
+      - For each requirement, report `OK` if satisfied or `FAIL` if not.
+      - Do not mark a requirement as `OK` without code evidence; for `OK` items provide only a compact pointer (file path + symbol + line range). For each requirement, provide a concise evidence pointer (file path + symbol + line range) excerpts only for `FAIL` requirements or when requirement is architectural, structural, or negative (e.g., "MUST NOT ..."). For such high-level requirements, cite the specific file paths or directory structures that prove compliance. Line ranges MUST be obtained from tooling output (e.g., `nl -ba` / `sed -n`) and MUST NOT be estimated. If evidence is missing, you MUST report `FAIL`. Do not assume implicit behavior.
+      - For every `FAIL`, provide evidence with a short explanation. Provide file path(s) and line numbers where possible.
+   - Perform a static analysis check by executing the `static-check` tool.
+      - Review the produced output and fix every reported issue in source code.
+      - Re-run the `static-check` tool until it produces no issues. If output is exactly `Error: no source files found in configured directories.`, treat it as successful no-source completion and continue without retries.
+   - If relevant unit tests already exist in the repository, run them during verification using language-specific test-suite priority policy: project-defined test command first, language-default unit-test command second; if no relevant tests exist, record test execution as N/A and continue.
+      - Verify that the implemented changes satisfy requirements evidence, static-analysis output, and unit-test outputs when tests are executed.
+      - If static analysis reports issues or executed unit tests fail, analyze whether they are caused by source defects or requirement-implementation mismatch. Assume requirement evidence is authoritative; when static analysis reports issues, fix source code unless requirements explicitly justify alternative handling.
+      - Fix the source code to resolve valid verification findings autonomously without asking for user intervention. Execute a strict fix loop: 1) analyze static-check output and unit-test failures (when tests ran), 2) determine root cause from evidence, 3) fix code, 4) re-run the `static-check` tool and re-run the selected unit-test suites when applicable. Repeat up to 2 times. If static analysis still reports issues after the second attempt, report the failure, OUTPUT exactly "ERROR: Refactor failed due to inability to complete static analysis!", delete the isolated worktree and branch with the `worktree-delete` tool, and then terminate the execution.
+      - Limitations: Do not introduce new features or change the architecture logic during this fix phase; if a fix requires substantial refactoring or requirements changes, report the failure, then OUTPUT exactly "ERROR: Refactor failed due to requirements incompatible with static-analysis constraints!", delete the isolated worktree and branch with the `worktree-delete` tool, and then terminate the execution.
+      - Do NOT create or modify tests in this repository.
+6. Update `%%DOC_PATH%%/WORKFLOW.md` via targeted edits using the canonical WORKFLOW.md contract (same terminology, same schema, same call-trace rules) and declaration file paths only, excluding line numbers, line ranges, and internal file-reference pointers.
+   - Update `%%DOC_PATH%%/WORKFLOW.md` as an LLM-first runtime model (English only) using a TARGETED EDIT policy.
+      - During generation/update, include declaration file paths only; MUST NOT include line numbers, line ranges, or internal file-reference pointers.
+      - Determine the change surface from repository evidence: run `git diff --name-only` and `git diff` to identify the modified files/symbols under %%SRC_PATHS%%.
+      - Modify ONLY the WORKFLOW.md sections impacted by those changes (execution unit index entries, execution unit subsections, and communication edges); preserve stable IDs and do not rewrite unrelated content.
+      - Ensure global consistency: if a changed internal symbol appears in any call-trace, update all affected call-trace nodes; if a unit/edge is added/removed, reflect it.
+   - Analyze only files under %%SRC_PATHS%% (everything else is out of scope) and identify ALL runtime execution units:
+      - OS processes (MUST include the main process).
+      - OS threads (per process), including their entry functions/methods.
+      - If no explicit thread creation is present, record "no explicit threads detected" for that process.
+   - For EACH execution unit, generate a complete internal call-trace tree starting from its entrypoint(s):
+      - Include ONLY internal functions/methods defined in repository source under %%SRC_PATHS%%.
+      - Do NOT include external boundaries (system/library/framework calls) as nodes; annotate them only as external boundaries where relevant.
+      - No maximum depth: expand until an internal leaf function or an external boundary is reached.
+   - Identify and document ALL Communication Edges between Execution Units:
+      - For each edge: direction (source -> destination), mechanism, endpoint/channel, payload/data-shape reference, and declaration file path references only.
+   - Preserve and maintain the canonical `WORKFLOW.md` schema:
+      - `## Execution Units Index`
+      - `## Execution Units`
+      - `## Communication Edges`
+   - Use stable execution unit IDs: `PROC:main`, `PROC:<name>` for processes; `THR:<proc_id>#<name>` for threads.
+   - Call-trace node format (MUST be consistent):
+      - `symbol_name(...)`: `<single-line role>` [`<defining filepath>`]
+         - `<optional: brief invariants/external boundaries>`
+         - `<child internal calls as nested bullet list, in call order>`
+7. Update `%%DOC_PATH%%/REFERENCES.md` references file
+   -  Create/update `%%DOC_PATH%%/REFERENCES.md` with the `references-generation` tool.
+8. **CRITICAL**: Stage & commit
+   - Show a summary of changes with `git diff` and `git diff --stat`.
+   - Stage changes explicitly (prefer targeted add; avoid `git add -A` if it may include unintended files): `git add <file...>` (ensure to include all modified source code & test and WORKFLOW.md only if it was modified/created).
+   - Ensure there is something to commit with: `git diff --cached --quiet && echo "Nothing to commit. Aborting."`. If command output contains "Aborting", OUTPUT exactly "No changes to commit.", and then delete the isolated worktree and branch with the `worktree-delete` tool, and then terminate the execution.
+   - Commit a structured commit message with: `git commit -m "refactor(<COMPONENT>)<BREAKING>: <DESCRIPTION> [useReq]"`
+      - Set `<COMPONENT>` to the most specific component, module, or function affected. If multiple areas are touched, choose the primary one. If you cannot identify a unique component, use `core`.
+      - Set `<DESCRIPTION>` to a short, clear summary in **English language** of what changed, including (when applicable) updates to: requirements/specs, source code, tests. Use present tense, avoid vague wording, and keep it under ~80 characters if possible.
+      - Set `<BREAKING>` to `!` if a breaking change was implemented (a modification to an API, library, or system that breaks backward compatibility, causing dependent client applications or code to fail or behave incorrectly), set empty otherwise.
+      - Include main features added, requirements changes, or a bug-fix description adding a multi-line comment (maximum 10 lines).
+         - Do not include the 'Co-authored-by' trailer or any AI attribution. A GPG-signed commit is not required.
+   - Confirm the repo is clean with the `git-status` tool. If the command returns an error code or prints any text containing "ERROR", override the final line with EXACTLY "WARNING: Refactor request completed with unclean git repository!".
+9.  **CRITICAL**: Merge Conflict Management
+   - Return to the original repository directory (the sibling directory of the worktree).
+   - Ensure you are on the original branch used before worktree creation by deriving `<BASE_PATH>` with the `base-path` tool if needed and executing `cd <BASE_PATH>`.
+   - Merge the isolated branch into the original branch: `git merge <WORKTREE_NAME>`
+   - If the merge completes successfully, delete the isolated worktree and branch with the `worktree-delete` tool; if the command returns an error code or prints any text containing "ERROR", OUTPUT exactly "ERROR: Worktree cleanup verification failed!", and then terminate the execution.
+   - If the merge fails or results in conflicts, do NOT remove the worktree directory and override the final line with EXACTLY "WARNING: Refactor request completed with merge conflicting!".
+10. Present results
+   - PRINT, in the response, the results for a human reader using clear, easily understandable sentences and readable Markdown formatting that highlight key findings, file paths, and concise evidence. Use the fixed report schema: ## **Outcome**, ## **Requirement Delta**, ## **Design Delta**, ## **Implementation Delta**, ## **Verification Delta**, ## **Evidence**, ## **Assumptions**, ## **Next Workflow**. Final line MUST be exactly: STATUS: OK or STATUS: ERROR.
+
+<h2 id="users-request">User's Request</h2>
+%%ARGS%%

package/src/resources/prompts/references.md

@@ -0,0 +1,100 @@
+---
+description: "Write a REFERENCES.md using the project's source code"
+argument-hint: "No arguments utilized by the prompt logic"
+usage: >
+  Select this prompt ONLY for docs-maintenance of %%DOC_PATH%%/REFERENCES.md, when that file is missing/outdated and you need to regenerate the repository navigation/index from evidence (entrypoints, modules, dependencies) and commit that doc change. Do NOT select if any other file (requirements, workflow, source, tests) must change; use /req-change, /req-new, /req-fix, /req-refactor, /req-cover, /req-implement, /req-create, or /req-recreate for those workflows. Do NOT select for read-only analysis/audits (use /req-analyze or /req-check).
+---
+
+# Write a REFERENCES.md using the project's source code
+
+## Purpose
+Maintain a machine-usable reference index (`%%DOC_PATH%%/REFERENCES.md`) derived from repository evidence so downstream LLM Agents MUST quickly discover entrypoints, modules, dependencies, and other navigational anchors during SRS-driven work.
+
+## Scope
+In scope: generate/update only `%%DOC_PATH%%/REFERENCES.md` in English using the `references-generation` tool and commit that doc change. Out of scope: changes to requirements, workflow docs, source code, or tests.
+
+## Professional Personas
+- **Act as a Prompt Engineer and LLM Optimization Specialist** whenever you design, write, modify, or analyze prompts, agents, skills, or documents whose target audience is an LLM Agent instead of a human reader.
+- **Act as a Senior System Engineer** when analyzing source code and directory structures to understand the system's architecture and logic.
+- **Act as a Technical Writer** when producing the final reference index, ensuring clarity, technical precision, and structured formatting.
+- **Act as an Expert GitOps Engineer** when executing git workflows, especially when creating/removing/managing git worktrees to isolate changes safely.
+
+## Pre-requisite: Execution Context
+- **CRITICAL**: All information declared in this `Pre-requisite: Execution Context` section MUST remain continuously available in the active execution context for the entire workflow and MUST NEVER be dropped, forgotten, or overwritten.
+- Generate <WORKTREE_NAME> with the `worktree-name` tool, retain the literal result for later steps, and use simple sequential execution with only linear shell commands compatible with restrictive filtering systems for all worktree operations in this workflow.
+
+
+## Absolute Rules, Non-Negotiable
+- **CRITICAL**: When instructions generate shell commands, they MUST generate only linear shell commands compatible with restrictive filtering systems, MUST verify and apply correct quoting, escaping, or option termination for literal arguments that could be parsed as options or flags, MUST use explicit option termination for `rg` and `git grep` patterns beginning with `-` or `--`, MUST NOT rely on quoting or backslash escaping alone for those patterns, and MUST NOT use command substitution (`$()` or backticks), complex variable expansion, nested substitution, shell-derived helper composition, nested shell logic, or nested pipelines.
+- **CRITICAL**: NEVER write, modify, edit, or delete files outside of the active git worktree directory, except under `/tmp`, and except for worktree operations executed through the `worktree-create` and `worktree-delete` tools.
+- You can read, write, or edit `%%DOC_PATH%%/REFERENCES.md`.
+- Treat static analysis as safe. Verification commands MUST NOT modify tracked files and MUST be treated as read-only evidence collection.
+- **CRITICAL**: Do not modify any project files except creating/updating `%%DOC_PATH%%/REFERENCES.md`.
+- **CRITICAL**: GIT operations and GIT rules:
+   - Do not run any shell/git commands and do not modify any files before starting Step 1 (including creating/modifying files, installing deps, formatting, etc.): **CRITICAL**: Check GIT Status.
+   - Step 1 may run only the git commands `git rev-parse --is-inside-work-tree`, `git rev-parse --verify HEAD`, `git status --porcelain`, and `git symbolic-ref -q HEAD` (plus minimal shell built-ins to combine their outputs into a single cleanliness check).
+   - If the repository is NOT clean (modified files, staged changes, OR untracked files), exit immediately without changing anything.
+   - At the end you MUST commit only the intended changes with a unique identifier and change description in the commit message.
+   - Leave the working tree AND index clean (git `status --porcelain` must be empty).
+   - Do NOT “fix” a dirty repo by force (no `git reset --hard`, no `git clean -fd`, no stash) unless explicitly requested. If dirty: abort.
+
+## Behavior
+- Do not perform unrelated edits.
+- If `.venv/bin/python` exists in the project root, use it for Python executions (eg, `PYTHONPATH=src .venv/bin/python -m <program name>`).
+- Non-Python tooling should use the project's standard commands.
+- Use filesystem/shell tools to read/write/delete files as needed (e.g., `cat`, `sed`, `perl -pi`, `printf > file`, `rm -f`, ...), but only to read project files and to write/update `%%DOC_PATH%%/REFERENCES.md`. Avoid in-place edits on any other path. Prefer read-only commands for analysis.
+
+
+## Execution Protocol (Global vs Local)
+You must manage the execution flow using two distinct methods:
+-  **Global Roadmap** (*check-list*): 
+   - You MUST maintain a *check-list* internally with `6` Steps (one item per Step).
+   - **Do NOT** use the *task-list tool* for this high-level roadmap.
+-  **Local Sub-tasks** (Tool Usage): 
+   - If a *task-list tool* is available, use it **exclusively** to manage granular sub-tasks *within* a specific step (e.g., in Step X: "1. Edit file A", "2. Edit file B"; or in Step Y: "1. Fix test K", "2. Fix test L").
+   - Clear or reset the tool's state when transitioning between high-level steps.
+
+## Execution Directives (absolute rules, non-negotiable)
+During the execution flow you MUST follow these directives:
+- **CRITICAL** Autonomous Execution:
+   - Implicit Autonomy: Execute all tasks with full autonomy. Do not request permission, confirmation, or feedback. Make executive decisions based on logic and technical best practices.
+   - Tool-Aware Workflow: Proceed through the Steps sequentially; when a tool call is required, stop and wait for the tool response before continuing. Never fabricate tool outputs or tool results. Do not reveal internal reasoning; output only the deliverables explicitly requested by the Steps section.
+   - Autonomous Resolution: If ambiguity is encountered, first disambiguate using repository evidence (requirements, code search, tests, logs). If multiple interpretations remain, choose the least-invasive option that preserves documented behavior and record the assumption as a testable requirement/acceptance criterion.
+   - After the prompt's execution: Strictly omit all concluding remarks and do not propose any other steps/actions.
+- **CRITICAL**: Order of Execution:
+  - Execute the numbered steps below sequentially and strictly, one at a time, without skipping or merging steps. Create and maintain a *check-list* internally while executing the Steps. Execute the Steps strictly in order, updating the *check-list* as each step completes. 
+- **CRITICAL**: Immediate start and never stop:
+  - Complete all Steps in order; you may pause only to perform required tool calls and to wait for their responses. Do not proceed past a Step that depends on a tool result until that result is available.
+  - Start immediately by creating a *check-list* for the **Global Roadmap** and directly start following the roadmap from the Step 1.
+
+
+## Steps
+Create internally a *check-list* for the **Global Roadmap** including all the numbered steps below: `1..6`, and start following the roadmap at the same time, executing the tool call of Step 1 (Check GIT Status). If a tool call is required in Step 1, invoke it immediately; otherwise proceed to Step 1 without additional commentary. Do not add extra intent-adjustment checks unless explicitly listed in the Steps section.
+1. **CRITICAL**: Check GIT Status
+   - Check GIT status with the `git-status` tool. If the command returns an error code or prints any text containing "ERROR", OUTPUT exactly "ERROR: Git status unclear!", and then terminate the execution.
+2. **CRITICAL**: Worktree Generation & Isolation
+   - Derive <BASE_PATH> with the `base-path` tool, derive <GIT_PATH> with the `git-path` tool, and generate <WORKTREE_NAME> with the `worktree-name` tool using simple sequential tool execution without shell composition.
+   - Create the dedicated isolated worktree with the `worktree-create` tool, then execute `cd <GIT_PATH>/../<WORKTREE_NAME>` before proceeding to the next step.
+   - If the command returns an error code or prints any text containing "ERROR", OUTPUT exactly "ERROR: Worktree generation failed!", and then terminate the execution.
+
+3. Update `%%DOC_PATH%%/REFERENCES.md` references file
+   -  Create/update `%%DOC_PATH%%/REFERENCES.md` with the `references-generation` tool.
+4. **CRITICAL**: Stage & commit
+   - Show a summary of changes with `git diff` and `git diff --stat`.
+   - Stage changes explicitly (prefer targeted add; avoid `git add -A` if it may include unintended files): `git add <file...>` (ensure to include only `%%DOC_PATH%%/REFERENCES.md`).
+   - Ensure there is something to commit with: `git diff --cached --quiet && echo "Nothing to commit. Aborting."`. If command output contains "Aborting", OUTPUT exactly "No changes to commit.", and then delete the isolated worktree and branch with the `worktree-delete` tool, and then terminate the execution.
+   - Commit a structured commit message with: `git commit -m "docs(<COMPONENT>)<BREAKING>: <DESCRIPTION> [useReq]"`
+      - Set `<COMPONENT>` to the most specific component, module, or function affected. If multiple areas are touched, choose the primary one. If you cannot identify a unique component, use `core`.
+      - Set `<DESCRIPTION>` to a short, clear summary in **English language** of what changed, including (when applicable) updates to: requirements/specs, source code, tests. Use present tense, avoid vague wording, and keep it under ~80 characters if possible.
+      - Set `<BREAKING>` to `!` if a breaking change was implemented (a modification to an API, library, or system that breaks backward compatibility, causing dependent client applications or code to fail or behave incorrectly), set empty otherwise.
+      - Include main features added, requirements changes, or a bug-fix description adding a multi-line comment (maximum 10 lines).
+         - Do not include the 'Co-authored-by' trailer or any AI attribution. A GPG-signed commit is not required.
+   - Confirm the repo is clean with the `git-status` tool. If the command returns an error code or prints any text containing "ERROR", override the final line with EXACTLY "WARNING: References request completed with unclean git repository!".
+5. **CRITICAL**: Merge Conflict Management
+   - Return to the original repository directory (the sibling directory of the worktree).
+   - Ensure you are on the original branch used before worktree creation by deriving `<BASE_PATH>` with the `base-path` tool if needed and executing `cd <BASE_PATH>`.
+   - Merge the isolated branch into the original branch: `git merge <WORKTREE_NAME>`
+   - If the merge completes successfully, delete the isolated worktree and branch with the `worktree-delete` tool; if the command returns an error code or prints any text containing "ERROR", OUTPUT exactly "ERROR: Worktree cleanup verification failed!", and then terminate the execution.
+   - If the merge fails or results in conflicts, do NOT remove the worktree directory and override the final line with EXACTLY "WARNING: References request completed with merge conflicting!".
+6. Present results
+   - PRINT, in the response, the results for a human reader using clear, easily understandable sentences and readable Markdown formatting that highlight key findings, file paths, and concise evidence. Use the fixed report schema: ## **Outcome**, ## **Requirement Delta**, ## **Design Delta**, ## **Implementation Delta**, ## **Verification Delta**, ## **Evidence**, ## **Assumptions**, ## **Next Workflow**. Final line MUST be exactly: STATUS: OK or STATUS: ERROR.