feat(core): enrich hardcoded system prompts in container templates by konard · Pull Request #190 · ProverCoderAI/docker-git

konard · 2026-03-24T09:06:38Z

Summary

Fixes #189

The system prompts injected into docker-git containers (CLAUDE.md, AGENTS.md, GEMINI.md at startup) were minimal — just agent identity and workspace context. The user requested enriching them based on all prompt-related issues.

What changed

New file: packages/lib/src/core/templates-entrypoint/system-prompt-content.ts

Shared behavioral instructions module — single source of truth for all agents
Contains enriched prompt sections extracted from issues

Modified template files (import and inject systemPromptBehavior):

packages/lib/src/core/templates-entrypoint/claude-extra-config.ts — Claude prompt
packages/lib/src/core/templates-entrypoint/codex.ts — Codex/AGENTS prompt
packages/lib/src/core/templates-entrypoint/gemini.ts — Gemini prompt

Modified test:

packages/app/tests/docker-git/entrypoint-auth.test.ts — updated assertions for new prompt structure

Docs (from previous commit):

CLAUDE.md and AGENTS.md — project-level instructions aligned with container prompts

New prompt sections (from issues)

Issue	Requirement	Added
#90	Isolated environment awareness	✅
#41	Role definition (mathematician-programmer)	✅
#41	Deep Research process rule	✅
#41	Rational comments format (CHANGE/WHY/REF)	✅
#41	Research loop on failures	✅
#41	Instrumental behavior (execute, don't guide)	✅
#101, #4	Mandatory subagent delegation + micro-agents	✅
#54	Plan mode on clone	✅
#96	Mandatory response contract (trust gap)	✅
#96	Claim safety rule (no "fixed" without proof)	✅
#96	Visibility rule (why user may not see result)	✅
#96	CI/CD rule (failing step + reproduce command)	✅
#92	Proof of execution in PRs (detailed)	✅
#78	Public API, no localhost URLs	✅

Architecture

The prompt has three layers:

Static code (system-prompt-content.ts → templates) — behavioral instructions injected at container startup
Dynamic (generated by templates-entrypoint/*.ts) — workspace context, permissions, tool access
Project-level (CLAUDE.md, AGENTS.md in repo root) — coding standards for this repo

This PR updates layer 1 (the code-embedded prompts) as requested.

Verification

npm run lint — 0 errors, 0 warnings
npm test — 192 tests pass (73 app + 119 lib)
All three agent templates now inject the enriched prompt
Main branch merged, no conflicts

Математические гарантии

Инварианты:

∀ agent ∈ {claude, codex, gemini}: prompt(agent) ⊇ systemPromptBehavior
Single source of truth: system-prompt-content.ts → all templates

Доказательства выполнения:

All 192 tests pass: verified locally after merge with main
Lint clean: 0 errors, 0 warnings
Code changes in template files (hardcoded prompts in code), not just docs

🤖 Generated with Claude Code

Adding .gitkeep for PR creation (default mode). This file will be removed when the task is complete. Issue: ProverCoderAI#189

Analyzed all prompt-related issues (ProverCoderAI#96, ProverCoderAI#101, ProverCoderAI#78, ProverCoderAI#41, ProverCoderAI#54, ProverCoderAI#92, ProverCoderAI#90, ProverCoderAI#109, ProverCoderAI#110, ProverCoderAI#111, ProverCoderAI#122, ProverCoderAI#4, ProverCoderAI#143, ProverCoderAI#108) and synthesized a comprehensive system prompt that incorporates: - Reasoning model from AGENTS.md (role simulation, not personal opinions) - Deep Research process rule (internal, not response format) - Instrumental behavior requirements (use tools, don't offer guides) - Mandatory subagent usage for task delegation (ProverCoderAI#101, ProverCoderAI#4) - Plan mode on clone (ProverCoderAI#54) - Mandatory response contract addressing trust gap (ProverCoderAI#96): status, root cause, what changed, verification, where to check - Proof of execution requirements in PRs (ProverCoderAI#92): screenshots, logs, artifacts in .knowledge/evidence/ - Research loop for uncertainty/failures - Stricter type safety rules (unknown/as boundaries) - Monadic composition bans (async/await, Promise chains in prod code) - Effect-TS resource management (acquireRelease + scoped) - Boundary data decoding requirement Both CLAUDE.md and AGENTS.md are now aligned with each other and with the more comprehensive version that was already in AGENTS.md, plus the new requirements from the issues listed above. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

This reverts commit c808bb5.

konard · 2026-03-24T09:22:03Z

🤖 Solution Draft Log

This log file contains the complete execution trace of the AI solution draft process.

💰 Cost estimation:

Public pricing estimate: $4.654270
Calculated by Anthropic: $2.908727 USD
Difference: $-1.745543 (-37.50%)

🤖 Models used:

Tool: Anthropic Claude Code
Requested: opus
Main model: Claude Opus 4.6 (claude-opus-4-6)
Additional models:
- Claude Haiku 4.5 (claude-haiku-4-5-20251001)

📎 Log file uploaded as Gist (1678KB)

View complete solution draft log

Now working session is ended, feel free to review and add any feedback on the solution draft.

konard · 2026-03-24T09:26:35Z

✅ Ready to merge

This pull request is now ready to be merged:

All CI checks have passed
No merge conflicts
No pending changes

Monitored by hive-mind with --auto-restart-until-mergeable flag

skulidropek · 2026-03-24T11:15:49Z

Ты не понял. Я хотел обновить системный прмот для docker-git контейнерах. ОН зашит в коде

konard · 2026-03-24T11:17:04Z

🤖 AI Work Session Started

Starting automated work session at 2026-03-24T11:17:02.908Z

The PR has been converted to draft mode while work is in progress.

This comment marks the beginning of an AI work session. Please wait for the session to finish, and provide your feedback.

Update the code-embedded system prompts that are injected into CLAUDE.md, AGENTS.md, and GEMINI.md when docker-git containers start. Extracts shared behavioral instructions into a new system-prompt-content.ts module and injects them into all three agent templates (claude, codex, gemini). New prompt sections based on issues ProverCoderAI#96, ProverCoderAI#101, ProverCoderAI#78, ProverCoderAI#41, ProverCoderAI#54, ProverCoderAI#92, ProverCoderAI#90: - Isolated environment awareness - Instrumental behavior (execute, don't guide) - Deep Research process rule - Mandatory subagent delegation with plan mode - Mandatory response contract (status, root cause, changes, verification) - Proof of execution requirements in PRs - Public API usage (no localhost URLs) Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

konard · 2026-03-24T11:55:18Z

🤖 Solution Draft Log

This log file contains the complete execution trace of the AI solution draft process.

💰 Cost estimation:

Public pricing estimate: $6.379617
Calculated by Anthropic: $4.165636 USD
Difference: $-2.213981 (-34.70%)

🤖 Models used:

Tool: Anthropic Claude Code
Requested: opus
Main model: Claude Opus 4.6 (claude-opus-4-6)
Additional models:
- Claude Haiku 4.5 (claude-haiku-4-5-20251001)

📎 Log file uploaded as Gist (1754KB)

View complete solution draft log

Now working session is ended, feel free to review and add any feedback on the solution draft.

konard · 2026-03-24T11:55:25Z

🔄 Auto-restart 1/3

Detected uncommitted changes from previous run. Starting new session to review and commit or discard them.

Uncommitted files:

?? ci-logs/

Auto-restart will stop after changes are committed or discarded, or after 2 more iterations. Please wait until working session will end and give your feedback.

…equirements Add missing behavioral sections from issues ProverCoderAI#96, ProverCoderAI#41, ProverCoderAI#92, ProverCoderAI#4, ProverCoderAI#54: - Role/goal definition (mathematician-programmer, ProverCoderAI#41) - Research loop on failures (ProverCoderAI#41, AGENTS.md) - Claim safety rule - no "fixed" without proof (ProverCoderAI#96) - Visibility rule - check why user may not see result (ProverCoderAI#96) - CI/CD rule - always return failing step + reproduce command (ProverCoderAI#96) - Rational comments format (CHANGE/WHY/REF, ProverCoderAI#41) - Micro-agent creation capability (ProverCoderAI#4) - Detailed proof-of-execution requirements (ProverCoderAI#92) Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

konard · 2026-03-24T12:10:35Z

🔄 Auto-restart 1/3 Log

This log file contains the complete execution trace of the AI solution draft process.

💰 Cost estimation:

Public pricing estimate: $3.236396
Calculated by Anthropic: $1.770646 USD
Difference: $-1.465750 (-45.29%)

🤖 Models used:

Tool: Anthropic Claude Code
Requested: opus
Model: Claude Opus 4.6 (claude-opus-4-6)

📎 Log file uploaded as Gist (2978KB)

View complete solution draft log

Now working session is ended, feel free to review and add any feedback on the solution draft.

konard · 2026-03-24T12:10:42Z

✅ Ready to merge

This pull request is now ready to be merged:

All CI checks have passed
No merge conflicts
No pending changes

Monitored by hive-mind with --auto-restart-until-mergeable flag

skulidropek · 2026-05-13T09:30:17Z

@coderabbitai review

coderabbitai · 2026-05-13T09:30:22Z

✅ Actions performed

Review triggered.

Note: CodeRabbit is an incremental review system and does not re-review already reviewed commits. This command is applicable only when automatic reviews are paused.

coderabbitai · 2026-05-13T09:31:36Z

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro Plus

Run ID: 340538e5-247d-4f57-a2f8-a00e495e0f4e

📥 Commits

Reviewing files that changed from the base of the PR and between f964400 and 55ea4b5.

📒 Files selected for processing (7)

AGENTS.md
CLAUDE.md
packages/app/tests/docker-git/entrypoint-auth.test.ts
packages/lib/src/core/templates-entrypoint/claude-extra-config.ts
packages/lib/src/core/templates-entrypoint/codex.ts
packages/lib/src/core/templates-entrypoint/gemini.ts
packages/lib/src/core/templates-entrypoint/system-prompt-content.ts

📝 Walkthrough

Summary by CodeRabbit

Release Notes

Documentation
- Updated agent operating instructions with stricter behavioral guidelines and mandatory response contracts.
- Enhanced architectural rules and process requirements for better consistency.
- Added explicit proof of execution documentation standards.
Refactor
- Made system prompt behavior configurable and dynamically injectable across all agent templates.
Tests
- Updated test assertions to verify new behavior blocks in agent prompts.

Walkthrough

This PR extracts shared behavioral instructions into a reusable constant, documents mandatory agent operating guidelines and architectural rules, and injects the behavior into Claude, Codex, and Gemini entrypoint templates via placeholder substitution. Test assertions verify behavior integration.

Changes

System Prompt Behavior and Agent Guidelines

Layer / File(s)	Summary
Shared system prompt behavior definition `packages/lib/src/core/templates-entrypoint/system-prompt-content.ts`	New `systemPromptBehavior` constant exports static plain-text instructions covering role/goal, mandatory rules (no browser, verify via builds/tests, real commands only), deep research guidelines, subagent delegation requirements, strict answer contract (Status, Root Cause, What changed, Verification, Where to check), safety/visibility rules, PR proof requirements, and inline code-comment templates.
Agent operating guidelines `AGENTS.md`	Response behavior wording refined; mandatory subagent usage added; strict response contract introduced; TypeScript `as`-casting rule clarified to allow only in one "axiomatic" module; Proof of Execution section added with minimum evidence categories and storage location (`.knowledge/evidence/<issue-or-pr>/...`); BEFORE MERGE checklist extended to require proofs attachment.
Architectural and coding rules `CLAUDE.md`	Reasoning model requirements and Functional Core/Imperative Shell principles clarified; Effect-based composition rules enforced (prohibition on `async/await`, Promise chains, `Promise.all`; mandated `Effect.try`/`Effect.tryPromise` for interop; `Effect.acquireRelease`/`Effect.scoped` for resources); requirement markup templates (QUOTE/REF/SOURCE) added; exhaustive pattern matching with `Match` required; Proof of Execution section with evidence types (UI/UX, API/Backend, Bugfix, Data/Migration, Performance) introduced; unit test example refactored from `async/await` to `it.effect(...)` with pipe composition, Effect.provide, and Effect.asVoid; quality checklists expanded with `async/await`, Promise chains, `try/catch`, and `console.*` prohibitions; proofs attachment required on BEFORE MERGE.
Claude entrypoint template integration `packages/lib/src/core/templates-entrypoint/claude-extra-config.ts`	`systemPromptBehavior` imported; global prompt template now uses `__SYSTEM_PROMPT_BEHAVIOR__` placeholder; `renderClaudeGlobalPromptSetup` performs `.replaceAll("__SYSTEM_PROMPT_BEHAVIOR__", systemPromptBehavior)`.
Codex/AGENTS entrypoint template integration `packages/lib/src/core/templates-entrypoint/codex.ts`	`systemPromptBehavior` imported; `__SYSTEM_PROMPT_BEHAVIOR__` placeholder introduced in template; `SUBAGENTS_LINE` replaced with `$BEHAVIOR_BLOCK` in both initial and update AGENTS.md templates; `renderEntrypointAgentsNotice` performs placeholder replacement.
Gemini entrypoint template integration `packages/lib/src/core/templates-entrypoint/gemini.ts`	`systemPromptBehavior` imported; GEMINI.md notice template updated with `__SYSTEM_PROMPT_BEHAVIOR__` placeholder replacing hard-coded instruction; `renderEntrypointGeminiNotice` performs placeholder replacement alongside existing `__GEMINI_HOME__` and `__TARGET_DIR__` substitutions.
Test verification of behavior injection `packages/app/tests/docker-git/entrypoint-auth.test.ts`	Rendered entrypoint assertions updated to check for `BEHAVIOR_BLOCK=` and Russian prompt block markers (`SUBAGENTS (ОБЯЗАТЕЛЬНО):`, `ОБЯЗАТЕЛЬНЫЙ КОНТРАКТ ОТВЕТА:`, `DEEP RESEARCH:`), replacing prior `SUBAGENTS_LINE="..."` string and occurrence-count assertions.

Estimated code review effort

🎯 3 (Moderate) | ⏱️ ~25 minutes

🐰 A prompt so shared, now lives everywhere,
Through Claude and Gemini, floating in air,
With proof and with purpose, each agent now knows
What must be done—onward the docker-git flow goes! ✨

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title accurately describes the main change: enriching hardcoded system prompts in container templates with shared behavioral instructions.
Description check	✅ Passed	The description provides comprehensive coverage including what changed, new prompt sections mapped to issues, architecture layers, and verification steps. It aligns well with the repository template.
Linked Issues check	✅ Passed	The PR fully addresses issue `#189` by synthesizing system prompt requirements from multiple prompt-related issues and embedding enriched behavioral instructions into container templates.
Out of Scope Changes check	✅ Passed	All changes directly support the main objective of enriching system prompts with behavioral instructions. Template modifications, new shared content module, and test updates are all in scope.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

🧪 Generate unit tests (beta)

Create PR with unit tests

⚔️ Resolve merge conflicts

Resolve merge conflict in branch issue-189-ba783568b6b3

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

konard and others added 2 commits March 24, 2026 09:06

Initial commit with task details

c808bb5

Adding .gitkeep for PR creation (default mode). This file will be removed when the task is complete. Issue: ProverCoderAI#189

konard marked this pull request as ready for review March 24, 2026 09:21

Revert "Initial commit with task details"

2fff509

This reverts commit c808bb5.

konard marked this pull request as draft March 24, 2026 11:17

konard changed the title ~~docs(prompt): unify system prompt from all issue requirements~~ feat(core): enrich hardcoded system prompts in container templates Mar 24, 2026

konard marked this pull request as ready for review March 24, 2026 11:54

konard and others added 2 commits March 24, 2026 12:04

Merge remote-tracking branch 'origin/main' into issue-189-ba783568b6b3

1f3a3e2

coderabbitai Bot approved these changes May 13, 2026

View reviewed changes

Conversation

konard commented Mar 24, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

What changed

New prompt sections (from issues)

Architecture

Verification

Математические гарантии

Инварианты:

Доказательства выполнения:

Uh oh!

konard commented Mar 24, 2026

🤖 Solution Draft Log

💰 Cost estimation:

🤖 Models used:

📎 Log file uploaded as Gist (1678KB)

Uh oh!

konard commented Mar 24, 2026

✅ Ready to merge

Uh oh!

skulidropek commented Mar 24, 2026

Uh oh!

konard commented Mar 24, 2026

Uh oh!

konard commented Mar 24, 2026

🤖 Solution Draft Log

💰 Cost estimation:

🤖 Models used:

📎 Log file uploaded as Gist (1754KB)

Uh oh!

konard commented Mar 24, 2026

🔄 Auto-restart 1/3

Uh oh!

konard commented Mar 24, 2026

🔄 Auto-restart 1/3 Log

💰 Cost estimation:

🤖 Models used:

📎 Log file uploaded as Gist (2978KB)

Uh oh!

konard commented Mar 24, 2026

✅ Ready to merge

Uh oh!

skulidropek commented May 13, 2026

Uh oh!

coderabbitai Bot commented May 13, 2026

Uh oh!

coderabbitai Bot commented May 13, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary by CodeRabbit

Release Notes

Walkthrough

Changes

Estimated code review effort

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

konard commented Mar 24, 2026 •

edited

Loading

coderabbitai Bot commented May 13, 2026 •

edited

Loading