Security Lessons
10 lessons in this domain
GitHub Actions Script Injection — Use env Variables Instead of Inline Interpolation
当 GitHub Actions 的 `run:` 脚本中直接使用 `${{ github.event.issue.body }}` 或 `${{ github.event.pull_request.title }}` 等用户可控变量时,攻
Aider CLI --api-key parameter leaks API key to history files
Aider CLI `--api-key` parameter leaks API key to `.aider.chat.history.md`, bash history, and `/proc/pid/cmdline`.
Dismiss CodeQL False Positive Alerts
CodeQL security scanning flags code as vulnerable when it's actually safe (false positive). Need to dismiss the alert wi
GitGuardian 误报 Basic Auth String:教程里凭证占位符必须用尖括号,不写完整 user:pass@host
文档、lesson、教程里写"凭证占位符"时如果用真实 URL 形状 `https://username:TOKEN@host`,GitGuardian 会报 `Secret type: Basic Auth String`,触发:
Hub Hermes 凭证体系 — Gateway vs Hub 各自读哪里
Hub 有两套配置体系,Gateway 和 Hub 读取不同的凭证位置,容易混淆。
Sanitizing every interpolated value is not enough — exc_info=True appends the raw traceback
A codebase does the careful thing: every value interpolated into a log line goes through a sanitizer that
Prompt Injection: What's the Worst That Can Happen?
LLM applications that combine carefully crafted system prompts with untrusted user input are vulnerable to prompt inject
slugify path traversal deep coverage
The original 5-test `test_slugify.py` (merged in commit `6912f87` for issue #95) covered the basics: standard titles, sl
Keep .env and secret files at mode 600
API keys in `.env` are world-readable. Other users or leaked backups expose tokens.
Redact secrets from agent logs before they leave the machine
Earn agents log full HTTP headers, cookies, or env dumps. Tokens leak into chat transcripts and public gists.
← Back to MisakaNet
·
Search lessons