Skip to content

Releases: TheStack-ai/swing-skills

v3.1.0 — Swing: 6 cognitive firewalls for Claude Code

28 Apr 02:54
@TheStack-ai TheStack-ai
v3.1.0
1c4d99f

Choose a tag to compare

View all tags
v3.1.0 — Swing: 6 cognitive firewalls for Claude Code Latest
Latest

Swing v3.1.0 — Open-Source Skills Framework for Claude Code

Your AI is not swinging at all. It picks the safe default, agrees when it should push back, cites no sources, hides its reasoning, and assumes everything will work. Six cognitive failures. Six firewalls.

What is Swing?

Swing is an open-source skills framework that adds 6 cognitive firewalls to Claude Code (and any AI agent via the Agent Skills open standard), preventing hallucination, confirmation bias, premature closure, anchoring, opaque reasoning, and optimism bias.

Install

npx skills add thestack-ai/swing-skills --all

Compatible with Claude Code, Codex CLI, Cursor (via Agent Skills).

The 6 Firewalls

Skill Cognitive Failure What It Forces
swing-clarify Premature closure 5W1H decomposition, ambiguity scoring, ≤3 questions before execution
swing-research Hallucination 4-stage verified research with S/A/B/C source tier grading
swing-options Anchoring bias 5 probability-weighted options including unconventional alternatives
swing-review Confirmation bias Steel-man + 3-vector attack. "Looks good" is structurally banned
swing-trace Black-box reasoning Assumption inventory, weakest-link analysis
swing-mortem Optimism bias Pre-mortem with 5 specific failure scenarios + circuit breakers

Concrete Before / After

The README shows real cases — including a baseline AI giving a wrong answer about SQLite concurrency that swing-research corrected with verified sources, and a code review where the baseline missed a Critical security vulnerability that swing-review caught.

What's Changed since v3.0.0

  • refactor: rebrand from "Stack Skills" to Swing for clarity around the AI-cognitive-failure framing
  • docs: add tagline "Your AI is not swinging at all"
  • fix: add Gotchas section to all 6 skills (flagged by pulser skill linter)

Compatibility

  • Claude Code (Anthropic)
  • Codex CLI (OpenAI)
  • Any AI agent supporting the Agent Skills standard

License

MIT — use freely, fork, contribute. PRs welcome.

Links

Assets 2
Loading