HISTORY.md

Project History and Knowledge Log

Every code, logic, or design change must be documented here with full context. Format: ## [Date] — vX.Y — [Module Changed]

---

Template for New Entries

## [YYYY-MM-DD] — vX.Y — [Module Name]
**Changed by:** [Name / Agent]
**File(s) affected:** [list of files]
**Reason:** [why was this change made]
**What changed:** [what exactly changed]
**Impact:** [what breaks or improves as a result]
**Linked issue:** [error log entry if applicable]

---

Version 0.1 — Project Inception

Date: 2026-04-07 (estimated) Changed by: Manas

Summary

Initial idea finalized.

• Project selected: Git Conflict Resolver
• Target platform: Antigravity OpenEnv
• Deployment target: Hugging Face Spaces
• Language: Python 3.10+
• Core concept: Autonomous AI agent that resolves Git merge conflicts inside a sandboxed repository, evaluated with a dense reward function
• Environment model: OpenEnv standard (reset / step / state API)

---

Version 0.2 — Task Design

Changed by: Manas

Summary

Created task design for the 3 core difficulty levels.

Task ID	Name	Type
Task 1	easy_conflict	Single-file conflict with standard markers
Task 2	medium_conflict	Multi-file dependency conflict — signature mismatch
Task 3	hard_conflict	Semantic merge conflict — logic duplication across branches

Files affected:

• CHECKLIST.MD — Phase 2 items defined
• ROADMAP.md — Task folders listed under "Task Repository Creation"

Impact: Defined scope of evaluation for the agent. All three tasks must be solvable before deployment.

---

Version 0.3 — Environment API Design

Changed by: Manas

Summary

Implemented environment API contract and reward function structure.

Files affected:

• skills/environment_rules.md — reset(), step(), state() API defined
• skills/grading_logic.md — Reward components and scoring table defined
• ROADMAP.md — Reward values documented

Reward structure introduced:

Reward Event	Value
Conflict marker correctly removed	+0.15 per block
File passes AST syntax check	+0.20
File fails AST parse	−0.10
All unit tests pass	+0.30
Duplicated logic detected	−0.20

Impact: Defined the learning signal for agent training. Reward is dense — given at every step, not just episode end.

---

Version 0.4 — Skills Documentation

Changed by: Manas

Summary

Documented the 5 agent skills.

Files affected:

• skills/agent_purpose.md — Primary objective and agent role defined
• skills/agent_skills.md — 5 skill modules described:
  1. Conflict Detection (marker scanning)
  2. Syntax Validation (AST-based)
  3. Cross-file Dependency Resolution
  4. Semantic Merge (logic deduplication)
  5. Test Validation (unit test execution)

Impact: Skills serve as both implementation guide and agent self-description for LLM-based agent prompting.

---

Version 0.5 — Documentation Structure

Changed by: Manas

Summary

Created docs/ directory with two key logging files.

Files created:

• docs/idea_and_development_chat.md — Architecture and design decision log
• docs/error_and_solving_chat.md — Runtime error and solution log

Purpose: Maintain a living record of design reasoning and debugging history throughout development. This prevents repeated mistakes and helps onboarding.

---

Update Rules (Mandatory)

Every change made to this project must be logged here. An entry is required when:

1. ✅ A new file is created
2. ✅ An existing file's logic changes
3. ✅ A reward value is adjusted
4. ✅ A skill definition changes
5. ✅ A task repository is modified
6. ✅ A bug is fixed (link to error_and_solving_chat.md)
7. ✅ A deployment is made

An entry must include:

• The date
• Which module/file changed
• Why it changed
• What the impact is
• Any linked issue or error