一键导入
radisha-improve-skill
Automatically detect and improve skills when gaps are noticed during work - proactively proposes updates
用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
菜单
Automatically detect and improve skills when gaps are noticed during work - proactively proposes updates
用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Given a TheRock nightly build (URL or run-id), return the rocm-systems pin_sha used in that build
Check whether a given rocm-systems commit is included in a specific TheRock nightly build
Find the first TheRock nightly build that includes a given rocm-systems commit
Reviews Pull Requests or local diffs with an 8-agent fan-out covering static analysis, dead code, code smells + quality (naming, complexity, single-responsibility, magic numbers), language rules (C++/Python/CMake), architecture, simplification, performance (hot-path classification, allocations, locks, I/O), and undefined behaviour (signed overflow, lifetime, strict aliasing, data races, sanitizer coverage; C/C++/unsafe-Rust only). Use when the user asks to "review this PR", "review the diff", "audit this branch", "/pr-review", or when staging changes before push.
Walk through a PR review interactively, one finding at a time. Generate review via pr-review, then for each issue present analysis + proposed inline comment, let user accept/edit/skip, accumulate into a PENDING GitHub review, submit at end.
Use when user wants to list, analyze, review, or summarize GitHub PR comments on a pull request number or URL
| name | radisha-improve-skill |
| description | Automatically detect and improve skills when gaps are noticed during work - proactively proposes updates |
Proactive skill improvement. When you notice a gap in a skill during work, this skill automatically:
DO NOT wait for the user to ask. Propose improvements immediately.
When you notice a gap, identify the skill by:
| Signal | Responsible Skill |
|---|---|
| Currently loaded skill (via Skill tool) | That skill |
| Agent prompt references a skill | That skill |
| Task type matches a skill category | See mapping below |
| Pattern/rule domain | Domain-specific skill |
| Task You're Doing | Likely Responsible Skill |
|---|---|
| PR review, code quality | pr-review, code-smells, static-analysis |
| C++ implementation | programming-cpp, programming-cpp-design-patterns |
| Python implementation | programming-python |
| CMake work | programming-cmake-best-practices |
| Planning features | planning-feature, planning-base |
| Architecture analysis | architecture-analyze |
| Writing tests (C++) | testing-gtest-gmock |
| Writing tests (Python) | testing-pytest |
| Git operations | git-commit, git-prepare-pull-request |
| AMD SMI code | library-amd-smi |
| Debugging profiler | debugging-rocprof-sys |
Automatically propose improvement when:
IF (using skill X) AND (any of these occur):
- You need a rule that doesn't exist
- The skill's threshold seems wrong for this case
- A pattern you encountered isn't documented
- You found a better way to do something
- The skill's guidance was incomplete
- You had to look up something externally
THEN:
- Invoke this skill
- Propose the improvement
┌─────────────────────────────────────────────────────────────────┐
│ AUTOMATIC: Gap Detected During Work │
│ "I need X but the skill doesn't cover it" │
└─────────────────────────────────────────────────────────────────┘
│
▼
┌─────────────────────────────────────────────────────────────────┐
│ AUTOMATIC: Identify Responsible Skill │
│ Use task-to-skill mapping + context │
└─────────────────────────────────────────────────────────────────┘
│
▼
┌─────────────────────────────────────────────────────────────────┐
│ AUTOMATIC: Draft Improvement Proposal │
│ Match skill style, find right section │
└─────────────────────────────────────────────────────────────────┘
│
▼
┌─────────────────────────────────────────────────────────────────┐
│ PROPOSE TO USER │
│ "I noticed X is missing from skill Y. Add this?" │
└─────────────────────────────────────────────────────────────────┘
│ │
no yes
│ │
▼ ▼
┌─────────┐ ┌───────────────────────┐
│Continue │ │ Apply + Report │
│ work │ └───────────────────────┘
└─────────┘
When proposing an improvement to the user, use this format:
---
**Skill Improvement Detected**
**Skill:** `[skill-name]`
**Gap:** [What's missing or wrong]
**Context:** [How you discovered this]
**Proposed Addition:**
[Show the exact content to add, in skill's style]
**Location:** [Section where it would go]
---
Apply this improvement? (y/n)
---
**Skill Improvement Detected**
**Skill:** `code-smells`
**Gap:** Missing "Parallel Inheritance Hierarchies" detection criteria
**Context:** During PR review, found this smell but skill didn't have clear thresholds
**Proposed Addition:**
### Parallel Inheritance Hierarchies
**Detection:**
- Every time you create a subclass of one class, you need to create a subclass of another
- Prefixes in class names match across hierarchies (e.g., `PhoneDevice`/`PhoneDialer`)
**Severity:** Should Fix (50)
**Location:** Category 3: Change Preventers section
---
Apply this improvement? (y/n)
Skills can be in multiple locations:
| Location | Path | Priority |
|---|---|---|
| Global skills | ~/.claude/skills/<skill-name>/SKILL.md | Check first |
| Radisha project | ~/work/radisha/skills/<skill-name>/SKILL.md | Primary source |
| Current project | ./skills/<skill-name>/SKILL.md | Project-specific |
To find a skill:
# Check global skills (symlinked from radisha)
ls -la ~/.claude/skills/<skill-name>/
# Find the actual source
readlink -f ~/.claude/skills/<skill-name>/SKILL.md
Always edit the source file in radisha project, not the symlink.
Before making changes:
---
name: skill-name
description: ...
---
# Skill Title
<IMPORTANT> key rules </IMPORTANT>
## When to Use
## Process/Phases
## Detailed Instructions
## Common Mistakes (if applicable)
## References
| Type | What to Add | Where |
|---|---|---|
| Missing rule | New check or guideline | Appropriate phase/section |
| Missing pattern | New detection criteria | Pattern/detection section |
| Better example | Code snippet or scenario | Examples section |
| Clarification | Explanation of edge case | Related rule section |
| New trigger | When skill should activate | "When to Use" section |
| Threshold adjustment | Different numeric value | Threshold/criteria section |
## [Section Name]
### [New Rule/Pattern Name]
**Why:** [Brief rationale]
**Detection:**
- [Criteria 1]
- [Criteria 2]
**Example:**
```code
// Bad
[bad example]
// Good
[good example]
Severity: [Critical/Must Fix/Should Fix/Nitpick]
## Phase 4: Apply Change
### Edit Checklist
- [ ] Found the source file (not symlink)
- [ ] Read surrounding context
- [ ] Matched existing style
- [ ] Added to appropriate section
- [ ] No contradictions with existing content
- [ ] Used Edit tool (not Write) for existing files
### Common Edit Locations
| Skill Type | Typical Sections to Update |
|------------|---------------------------|
| Programming skills | Rules tables, code examples |
| Planning skills | Phase checklists, task lists |
| Review skills | Detection criteria, severity tables |
| Debugging skills | Troubleshooting steps, symptoms |
## Phase 5: Report
After making the change, report:
```markdown
## Skill Improved: [skill-name]
**What was missing:** [Description of the gap]
**What was added:**
- [Summary of addition]
**Location:** [File:Line or Section name]
**Reason:** [Why this improvement helps]
Situation: Reviewing C++ code, you encounter a pattern that looks wrong but isn't in code-smells.
Internal thought: "This is a 'Refused Bequest' - subclass doesn't use parent methods. The skill mentions it but has no detection criteria."
Automatic action:
---
**Skill Improvement Detected**
**Skill:** `code-smells`
**Gap:** "Refused Bequest" lacks detection criteria
**Context:** Found during PR review of inheritance hierarchy
**Proposed Addition:**
**Detection criteria:**
- Subclass overrides parent method to do nothing or throw
- Subclass uses <30% of inherited interface
- Subclass only uses parent for type compatibility
**Severity:** Should Fix (50) - indicates wrong inheritance
**Location:** Object-Orientation Abusers section
---
Apply this improvement? (y/n)
Situation: Writing C++ code, you need a pattern for optional parameters but programming-cpp doesn't cover it.
Internal thought: "I know std::optional is the right choice here, but the skill doesn't mention it."
Automatic action:
---
**Skill Improvement Detected**
**Skill:** `programming-cpp`
**Gap:** No guidance on optional parameters
**Context:** Needed while implementing function with optional config
**Proposed Addition:**
### Optional Parameters
| Pattern | Use Case |
|---------|----------|
| `std::optional<T>` | Value may or may not be present |
| Default argument | Always has a sensible default |
| Overloads | Different behavior, not just presence |
```cpp
// Good: explicit optionality
void configure(std::optional<Config> config = std::nullopt);
// Bad: using pointer for optionality
void configure(Config* config = nullptr);
Location: Parameters section
Apply this improvement? (y/n)
### Example 3: After Debugging
**Situation:** Spent time debugging a rocprofiler-systems issue. The solution should be in the debugging skill.
**Internal thought:** "I figured out that `ROCPROFSYS_VERBOSE=3` helps diagnose this. Others will hit this too."
**Automatic action:**
```markdown
---
**Skill Improvement Detected**
**Skill:** `debugging-rocprof-sys`
**Gap:** Missing verbose logging troubleshooting step
**Context:** Discovered while debugging instrumentation failure
**Proposed Addition:**
### Enable Verbose Logging
When standard troubleshooting doesn't reveal the issue:
```bash
export ROCPROFSYS_VERBOSE=3
rocprof-sys-run -- ./app
Levels: 1 (errors), 2 (warnings), 3 (info), 4+ (debug)
Location: General Troubleshooting section
Apply this improvement? (y/n)
## When NOT to Use This Skill
| Situation | What to Do Instead |
|-----------|---------------------|
| Creating a new skill | Use `radisha-create-skill` |
| Major skill rewrite | Plan it as a feature |
| Project-specific rules | Add to project's CLAUDE.md |
| One-time exception | Just note it in the review |
## Quick Reference
```bash
# Find skill source
readlink -f ~/.claude/skills/<name>/SKILL.md
# Edit skill
# Use Edit tool on the source file
# After editing, changes are immediately available
# (skills are symlinked, no need to "deploy")
This skill works with:
radisha-create-skill - for new skillsradisha-update - for updating radisha itselfradisha-skills - to list available skills