dsh-skill-pack

93 original, field-tested skills for DeepSeek Harness agents — one command installs them all into your agent's skill folder.

Built from months of real agent operations: these skills encode the disciplines that actually caught our agent's mistakes — thinking frameworks that fire on the right triggers, verification workflows that refuse fake completion, memory discipline that survives context compaction.

Install

npm install -g dsh-skill-pack
dsh-skill-pack install          # → ~/.dsh/skills (skips existing unless --force)
dsh-skill-pack install --force  # overwrite same-name skills

Restart DSH after install (skills are scanned at startup).

What's inside

| Group | Count | Examples | |---|---|---| | Thinking frameworks | 28 | thinking-bounded-rationality, thinking-red-team, thinking-cynefin — trigger-conditioned reasoning skills | | Security assessment (dsh-sec-*) | 42 | dsh-sec-code-audit, dsh-sec-reverse-engineering, dsh-sec-llm-security, ... — authorized testing only; every skill carries a "for authorized..." scope line | | Verification & agent hygiene | 22 | dsh-verification, dsh-tdd, dsh-debugging, dsh-memory, dsh-self-evolution, dsh-grilling, ... |

Full inventory: skills/ — every directory is one skill (SKILL.md + assets).

Scope note

This pack contains original, self-authored skills only. Third-party skills (Anthropic official, obra/superpowers ecosystem, document-generation skills) are intentionally not included — install those from their own sources with their own licenses.

Security skills (dsh-sec-*) are defensive/offensive security-education content for authorized engagements only (pen-testing, CTF, malware analysis, reverse engineering). Each skill's description states its authorized-use scope. You are responsible for using them lawfully.

For maintainers

npm run build    # re-copy from ~/.dsh/skills (exclusion list in scripts/build-release.mjs) + frontmatter gate
node scripts/check-frontmatter.mjs

Provenance scan: node scripts/scan-provenance.mjs flags third-party markers in the source ~/.dsh/skills tree (both scripts live in this repo under scripts/).

Changelog

v1.0.1 (2026-09-02)

• Fix: package description and README counts corrected 94 → 93 (dsh-theme-factory, a third-party-origin skill, was excluded after the initial count)
• Description now names the security-assessment block explicitly

v1.0.0 (2026-09-02)

• First release: 93 skills (28 thinking-* + 42 dsh-sec-* + 22 agent-hygiene dsh-* + autotelic-evolution)
• Installer: additive-only copy (never deletes target skills), --force for overwrite
• Provenance vetted: 30 third-party-origin skills + 1 nested third-party project (src-hunter) excluded
• Gitleaks clean: 1 doc-example finding (curl auth header) fixed at source before release