// Report issues and propose fixes to improve your own capabilities when you encounter errors or limitations
| name | self-improve |
| description | Report issues and propose fixes to improve your own capabilities when you encounter errors or limitations |
You have access to gh CLI and git in the sandbox. Use this skill when
you encounter bugs, errors, or limitations in your own capabilities during
your work.
Recognize these situations during your normal work:
Always ask the user for consent before filing an issue or creating a PR. Briefly explain what you found and what you want to report, then ask for approval. Do NOT file anything without explicit user confirmation.
Do NOT derail the user's current task. Keep the consent request brief, file the issue quickly after approval, then continue with the user's request.
GITHUB_TOKEN env var is pre-configured (injected from GITHUB_BOT_TOKEN)git config neededUse when: you encounter any issue with your own capabilities. Time: <30 seconds. Don't interrupt the user's flow.
Template:
gh issue create --repo "ginlix-ai/LangAlpha" \
--title "bug(agent): <what broke>" \
--label "agent-reported" \
--body "<structured body>"
Issue body structure:
## What I was doing
<user's task context — what were you trying to accomplish>
## What went wrong
<exact error message or unexpected behavior>
## Where the issue likely is
<file paths, function names, skill names — be specific>
## Suggested fix
<if obvious, describe; otherwise "Needs investigation">
## Environment
- Thread: <thread_id if available>
- Tool/Skill: <which tool or skill was involved>
- Error type: <tool_error | skill_instruction | mcp_data | prompt | sandbox>
@claude Please triage this issue — verify the root cause, assess severity, and suggest a fix if straightforward.
Default to Workflow 1 (filing an issue). Only create a PR when the user explicitly asks you to fix it yourself. Do NOT propose PRs on your own initiative.
Steps:
.self-improve/langalpha exists, cd .self-improve/langalpha && git checkout main && git pull origin main to get latest. Otherwise gh repo clone "ginlix-ai/LangAlpha" .self-improve/langalpha -- --depth 1cd .self-improve/langalpha && git checkout main && git checkout -b bot/fix/<short-desc>ruff check . && pytest (or relevant subset)fix(scope): descriptiongh pr create --repo "ginlix-ai/LangAlpha" \
--base main \
--title "fix(agent): <what's fixed>" \
--label "agent-reported" \
--body "<structured body>"
PR body structure:
## Problem
<link to issue if filed, or describe the bug>
## Root Cause
<what was wrong and why>
## Fix
<what was changed and why this approach>
## Testing
<what tests were run, what was verified>
## Context
- Discovered during: <brief user task description>
- Thread: <thread_id>
Use this to identify the right module when filing issues or proposing fixes.
| Directory | What lives here | Example issues |
|---|---|---|
skills/ | Skill SKILL.md instructions and assets | Wrong examples in skills/dcf-model/SKILL.md, bug in a provided script snippet, outdated API usage, missing steps in a workflow, new best practice to add |
mcp_servers/ | MCP server implementations (yfinance, fundamentals, macro, price_data) | yfinance_mcp_server.py returns malformed data, a fundamentals endpoint is missing a field, macro data has wrong units |
src/tools/ | External tool implementations (web fetch, crawl, search, SEC, market data) | fetch.py times out on certain URLs, SEC filing parser fails on 10-K amendments, search returns stale results |
src/ptc_agent/agent/tools/ | Core sandbox tools (ExecuteCode, Bash, file ops, grep, glob, think, todo) | code_execution.py mishandles large stdout, bash.py doesn't escape special chars, file_ops.py fails on binary files |
src/ptc_agent/agent/middleware/ | Middleware stack (skills, subagents, plan mode, compaction, memory, caching) | Skill loading fails silently, subagent doesn't inherit context, compaction truncates important content |
src/ptc_agent/agent/prompts/ | System prompt templates (Jinja2) and config | Redundant or wrongful instructions in system.md.j2, useful tips and experience worth persisting into prompts |
agent-reported labelbug for broken behavior, enhancement for capability gapsskills, tools, mcp, prompt, sandboxmain — always bot/fix/ or bot/feat/ branchesmain branch contains the latest code. Always branch from main, target PRs to main.self-improve/langalpha (inside workspace, persists across restarts)Go through EVERY item before running gh issue create or gh pr create:
ginlix-ai/LangAlphamain, PR base is mainManage user profile including watchlists, portfolio, and preferences.
DCF valuation: free cash flow projections, WACC, terminal value, sensitivity analysis
Inline HTML widgets: charts, dashboards, data tables rendered directly in the chat via ShowWidget
Web scraping with Scrapling: MCP tool wrappers for quick fetching, plus direct Python API for advanced scraping with selectors, sessions, and spiders
Use this skill any time a spreadsheet file is the primary input or output. This means any task where the user wants to: open, read, edit, or fix an existing .xlsx, .xlsm, .csv, or .tsv file (e.g., adding columns, computing formulas, formatting, charting, cleaning messy data); create a new spreadsheet from scratch or from other data sources; or convert between tabular file formats. Trigger especially when the user references a spreadsheet file by name or path — even casually (like "the xlsx in my downloads") — and wants something done to it or produced from it. Also trigger for cleaning or restructuring messy tabular data files (malformed rows, misplaced headers, junk data) into proper spreadsheets. The deliverable must be a spreadsheet file. Do NOT trigger when the primary deliverable is a Word document, HTML report, standalone Python script, database pipeline, or Google Sheets API integration, even if tabular data is involved.
Search X (Twitter) posts, pull user profiles, fetch specific tweets, and read reply threads for sentiment, news, and event research. Triggers on 'X', 'Twitter', 'tweets about', 'sentiment on', 'what are people saying about', 'historical tweets', or any request to read public X content.