Skip to content

Releases

sstack 1.1.0

sstack 1.1.0 Latest
Latest

Choose a tag to compare

View all tags
@GeiserX GeiserX released this 07 Nov 15:13
v1.1.0
9054e92
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.
  • Renamed from agent-skills to sstack (sergio stack). GitHub redirects the old repository URLs; the documentation site moved to https://geiserx.github.io/sstack/.
  • Installable as a Claude Code plugin: /plugin marketplace add GeiserX/sstack, then /plugin install sstack@sstack. Plugin skills are namespaced, as in /sstack:file-pr.
  • scripts/install.sh links the skills (and with --hooks the sergio-loop runtime) in one command, replacing the installer that was pasted into the docs. It now also links sergio_loop_registration.py, which the SessionStart hook imports.
  • New skills: full-loop (sergio-loop to done, then refine-loop) and github-video-embed (videos that play inline on GitHub).
  • file-pr no longer requests reviewers and keeps provenance in one References section.
  • eli5, help-decide, docs-loop and refine-loop trigger on the user rather than one named person.
  • The runtime's default accounts folder is now ~/.config/sstack/accounts (SERGIO_LOOP_ACCOUNTS_DIR still overrides it). Point it at your accounts folder, or set the variable, before upgrading.
  • The validator checks that every skill has a README row and that the plugin and marketplace versions match.
Assets 2
Loading

v1.0.0

v1.0.0

Choose a tag to compare

View all tags
@GeiserX GeiserX released this 29 Sep 22:26
v1.0.0
b2f042f
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

First release as agent-skills, formerly claude-skills. The skills from claude-code-parallel-skills now live here too, and every skill comes in a Claude Code copy and a Codex copy.

Skills

  • Parallel: investigate, review-pr, review-code, research, implement. Independent agents run in parallel, the panel scales to the task, and every finding is checked before anyone acts on it. They no longer need oh-my-claudecode, and the command names dropped the !.
  • Knowledge base: kb-research and kb-review answer and review from your team's own record. You fill in one adapter file for whatever you have: a markdown folder, SQLite FTS5, a RAG endpoint, an MCP server or a CLI.
  • Loops: sergio-loop, refine-loop, docs-loop, unchanged.

Codex

codex/skills/ holds a Codex copy of every skill. Each one runs only when you name it ($review-pr) and says how parallel agents, model tiers and slash commands map to Codex. scripts/build_codex.py generates the copies, and a test fails when they drift from skills/.

Other people's skills

vendor/ holds Lauren Tan's pstack (via Michael Denyer's Claude Code port) and Matt Pocock's writing-for-agents, each with its licence and a pinned provenance note.

Install for Claude Code or Codex: docs/installation.md.

Loading