ffss — the flipflopslopstack.

uinaf/ffss

ffss, the flipflopslopstack. A state machine that doesn't believe the agent, a reviewer that doesn't like the agent, and some skills so the agent behaves. Ships slop; checks receipts.

CLIs

CLI	What it is
`slopmachine`	The deterministic state machine: Go CLI + SQLite spine gating intake → release → build → verify → review → deliver → watch, with forge-verified evidence
`slopguard`	Pre-ship second-model review closeout CLI

Skills

Install all nine as one Agent Plugins package. Claude Code, Codex, Cursor, and Grok read the same marketplace and the portable root manifest:

# Claude Code
/plugin marketplace add uinaf/ffss
/plugin install ffss@ffss

# Codex CLI
codex plugin marketplace add uinaf/ffss
codex plugin add ffss@ffss

# Cursor CLI (then /plugins in interactive mode to install)
cursor-agent plugin marketplace add https://github.com/uinaf/ffss

# Grok CLI
grok plugin install uinaf/ffss --trust

OpenCode plugins are JS hooks and custom tools, not a skill carrier; its agent skills are a separate native feature discovered from ~/.config/opencode/skills. Link the plugin's skills there, for example from the Claude Code checkout:

mkdir -p ~/.config/opencode/skills
ln -sfn ~/.claude/plugins/marketplaces/ffss/skills/* ~/.config/opencode/skills/

The plugin is SHA-versioned; updates follow this repo's main.
Skills marked disable-model-invocation load only on explicit /name invocation in harnesses that support the flag; Codex mirrors that policy with policy.allow_implicit_invocation: false in each matching agents/openai.yaml.
The repo root plugin.json is the shared manifest from the Agent Plugins standard.

One skill per lane; route by the artifact, not the verb:

Skill	Owns
`slopmachine`	The governed multi-unit run, driving the slopmachine CLI
`slopguard`	One independent code review, through the slopguard CLI
`slopcourier`	Opening the change request, with the visual-evidence ladder
`slopnanny`	Walking an open change request to a settled outcome
`slopclean`	AI tells: stripping slop from writing, a diff, or a test suite
`slopspec`	Tracker artifacts: issues, epics, durable work plans
`slopscriber`	Repo docs: README, AGENTS.md, docs/, specs; content, currency, house style
`slopprep`	Agent readiness: boot, gates, proof paths
`wat`	Chat replies: terse labeled deltas

The GitHub shell (PR/issue templates, SECURITY.md, rulesets, releases) is gh-setup's lane, in agent-skills. The one seam worth naming: AGENTS.md orientation and proof paths belong to slopprep; its compression, currency, and style belong to slopscriber.

Reference bindings elsewhere in the family: slopzapper (forge review bot), slopscouter (QA, planned), slopbench (evals), and slopwake.

License

MIT; see LICENSE; members carry their own copies.