RFC-0006 Quests
rfcs/RFC-0006-quests.md
- Feature Name:
quests - Start Date: 2026-11-10
- RFC PR: #29
- Updated by: N/A
To coordinate open collaboration across Recursive Self-Improvement (RSI) benchmarks, cross-harness evaluations, algorithm reproductions, and harness plugins, Shared RSI Community specifies an Issue-Template-driven Task Board (Quests) automated by rsi_community_steward.
1. Task Publishing Authority & Source of Truth
- Only Committee-Published Tasks Are Recognized: The community only recognizes tasks published by members of the RSI Community Committee. Task issues opened directly by non-committee users are not recognized, are ignored by the automated steward agent, do not appear on the Task Board, and do not count toward EXP or monthly contributor rewards.
- GitHub Issue as the Single Source of Truth: Every task — whether
open,in_progress, orcompleted— must have a corresponding standardized GitHub Issue inPrism-Shadow/penguin-harness-communityas its authoritative data source. - Proposing Task Ideas: Community members who wish to suggest a new task may share their proposal with the Committee's Proposal Group in the community WeChat group or during the bi-weekly meeting; once approved, a committee member will publish the official task issue.
2. Domain Groups, Sorting & Pagination
All recognized tasks are indexed in community/tasks.yaml of penguin-harness-community and rendered on /quests/ across four domain sections, each containing three equal-width progress columns:
- Four Domain Groups (
Domain):Benchmark: Standardized benchmark integrations (following RFC-0001, e.g., Integrate SWE-bench #30, Integrate FDE-gym #31, and completed tasks #39–#41).Evaluation: Cross-harness and frontier base-model self-evolution evaluations (following RFC-0002, e.g., Codex Self-Evolution Comparison #32, GPT-6 Evolution Capability Experiment #33, and completed tasks #42–#43).RSI Algorithm: Core RSI paper reproductions and plugin integrations (following RFC-0003, e.g., Integrate RSIAgent Method #34, Integrate RRSI Method #35).PenguinHarness: Core harness and plugin development (following RFC-0003, e.g., Develop Knowledge Base Plugin #36, Develop Ontology Plugin #37, Optimize Context Compaction #44, and High-Playability Self-Evolution Case Study #45).
- Three Equal-Width Progress Columns:
Open (open)In Progress (in_progress)Completed (completed). - Newest-First Ordering & Per-Column Pagination: Within each domain and progress column, tasks are ordered by timestamp descending (newest first) and paginated so individual sections stay compact.
- Cards & Detail Modal: Each standard task awards 50 EXP. Board cards show the title, prominent
+50 EXPbadge, Issue number (plus PR number when applicable), and assignee; clicking a card opens a modal showing the full task description, target repository, task issue, pull request, related issue, and plain URLs to original papers and repositories.
3. Standard GitHub Issue Format Specification
Committee members create and maintain task issues using .github/ISSUE_TEMPLATE/task.yml. Tasks across all three stages (open, in_progress, completed) share the same 12-section Markdown body schema.
3.1 Issue Title, Labels & State
- Issue Title: Must start with
[Task]followed by the English task title (English is the default language), for example:[Task] Integrate SWE-bench - Issue Labels: Must include the
tasklabel. - GitHub Issue State:
openandin_progresstasks remain Open on GitHub.completedtasks must be closed as Closed as completed.
3.2 12-Section Issue Body Structure (Matching .github/ISSUE_TEMPLATE/task.yml)
GitHub Issue Forms renders the template into 12 level-3 Markdown headings (###):
### Domain
Benchmark
### Status
open
### EXP
50
### Target Repository
https://github.com/Prism-Shadow/penguin-harness-benchmark
### Assignee
_No response_
### Pull Request
_No response_
### Related Issue
_No response_
### Title (English)
Integrate SWE-bench
### Title (Chinese)
集成 SWE-bench
### Description (English)
Integrate the SWE-bench software engineering benchmark into `penguin-harness-benchmark` with standardized container environments and grading scripts (per RFC-0001).
### Description (Chinese)
将真实 GitHub 软件工程评测基准 SWE-bench 接入 `penguin-harness-benchmark`,封装标准化容器评测环境与自动评分脚本(遵循 RFC-0001 规范)。
### Original Papers & Repository Links
https://github.com/SWE-bench/SWE-bench
https://arxiv.org/abs/2310.06770
3.3 Stage-Specific Requirements (open / in_progress / completed)
Every task issue must satisfy the stage-specific requirements for ### Status, ### Assignee, ### Pull Request, and GitHub Issue state shown below:
| Task Stage | GitHub Issue State | ### Status |
### Assignee |
### Pull Request |
Live Examples |
|---|---|---|---|---|---|
Open (open) |
OPEN |
open |
_No response_ |
_No response_ |
#30, #45 |
In Progress (in_progress) |
OPEN |
in_progress |
@<github-handle> (required, e.g., @hiyouga) |
Plain URL to the active PR in the target repository (e.g., https://github.com/Prism-Shadow/penguin-harness/pull/1491, or _No response_ if not yet opened) |
#44 |
Completed (completed) |
CLOSED (completed) |
completed |
@<github-handle> (required, e.g., @Saigyouji-Yuyuko1000) |
Plain URL to the completed PR in the target repository (required, e.g., https://github.com/Prism-Shadow/penguin-harness-benchmark/pull/16) |
#39–#43 |
3.4 Field Parsing & Single-Language Fallback Rules
Section Heading (###) |
Requirement | Format & Fallback Behavior |
|---|---|---|
### Domain |
Required | One of Benchmark, Evaluation, RSI Algorithm, or PenguinHarness. |
### Status |
Required | One of open, in_progress, or completed. |
### EXP |
Required | Positive integer (default 50) specifying the experience points awarded upon completion. |
### Target Repository |
Required | Target repository URL where the Pull Request must be submitted; must be one of the four official repositories: 1. https://github.com/Prism-Shadow/penguin-harness2. https://github.com/Prism-Shadow/penguin-harness-community3. https://github.com/Prism-Shadow/penguin-harness-benchmark4. https://github.com/Prism-Shadow/penguin-harness-plugins |
### Assignee |
Required for in_progress & completed |
Contributor's GitHub handle prefixed with @ (e.g., @hiyouga); _No response_ when open. |
### Pull Request |
Required for completed |
Plain URL to the Pull Request submitted to the target repository; _No response_ when not yet opened. |
### Related Issue |
Optional | Optional related GitHub Issue plain URL (_No response_ if none). |
### Title (English) |
Either EN or ZH required | Task title in English (default language). If blank or _No response_, automatically falls back to ### Title (Chinese). |
### Title (Chinese) |
Either EN or ZH required | Task title in Chinese. If blank or _No response_, automatically falls back to ### Title (English). |
### Description (English) |
Either EN or ZH required | Task description and acceptance criteria in English. If blank or _No response_, automatically falls back to ### Description (Chinese). |
### Description (Chinese) |
Either EN or ZH required | Task description and acceptance criteria in Chinese. If blank or _No response_, automatically falls back to ### Description (English). |
### Original Papers & Repository Links |
Optional | External reference papers or upstream code repositories — one plain http:// or https:// URL per line (no custom titles). |
4. Claiming, Status Transition Comments & Completion Settlement
- Transitioning to
in_progress(/claim):- Contributors claim an open task by replying
/claimunder its official GitHub Issue (or/claim @<handle> <pr-url>if a PR is already open). - Concurrent Claim Limit & Completion Grace Period: Each GitHub user may hold at most 2
in_progresstasks at the same time; additional/claimrequests beyond 2 are ignored. Tasks have a 1-month (30-day) completion grace period — claims not completed within 1 month (or released via/unclaim) automatically return toopen. - Once verified,
rsi_community_stewardupdates the Issue body's### Statustoin_progressand### Assigneeto@<handle>(plus### Pull Requestif provided).
- Contributors claim an open task by replying
- Transitioning to
completed(/complete):- Once the Pull Request in
### Target Repositoryis reviewed and merged by the Committee's Review Group, a standardized completion comment is recorded on the task Issue:/complete @<github-handle> <pr-url> (+50 EXP) rsi_community_steward(or a committee member) updates the Issue body's### Statustocompleted,### Assigneeto@<github-handle>, and### Pull Requestto<pr-url>, and closes the Issue ascompleted.- The task's EXP is immediately settled onto the Monthly and All-Time Leaderboards (see RFC-0007 People & Rewards).
- Once the Pull Request in