Problem
The agent currently runs with minimal context about the repository it's working on. It needs a proper system prompt that gives it repo-specific knowledge.
Expected behavior
- Agent receives a rich system prompt including:
- Repository README (or summary)
- Recent commit history / changelog
- Open issues and their status
- PR context (if responding to a PR)
- Repo conventions (language, framework, testing approach)
- Custom instructions from
.openclaw.yml config file
Implementation notes
- Fetch README.md content via GitHub API
- Fetch recent commits (
git log --oneline -20)
- Fetch open issues list (titles + labels)
- Support a
.openclaw.yml or .github/openclaw.yml config file for custom instructions:
system_prompt: "You are a helpful code reviewer focused on TypeScript best practices"
context:
include_readme: true
recent_commits: 20
open_issues: true
- Be mindful of context window limits — summarize large READMEs
- Cache context between runs (use Actions Cache)
Priority
P1 — Agent quality depends heavily on context. Without this, responses will be generic.
Problem
The agent currently runs with minimal context about the repository it's working on. It needs a proper system prompt that gives it repo-specific knowledge.
Expected behavior
.openclaw.ymlconfig fileImplementation notes
git log --oneline -20).openclaw.ymlor.github/openclaw.ymlconfig file for custom instructions:Priority
P1 — Agent quality depends heavily on context. Without this, responses will be generic.