Fernando RamosGuadalajara --:--
← Back to home

luci-efe/skills · Open-source suite

Engineering skills with evidence.

Agents are not chat windows. These nine skills turn the Six Rules of Agentic Engineering into executable decision trees, explicit tool checks, and revision-bound evidence across Claude Code, Codex, and Antigravity.

Install the full suitegithub.com/luci-efe/skills ↗
npx skills@latest add luci-efe/skills

Claude Code plugin
claude plugin marketplace add luci-efe/skills && claude plugin install ramos-skills@ramos

Then run /doctor in your repo. State lives in .ramos/ (override with RAMOS_STATE_DIR). With hooks installed, a passing /proof run in a Claude Code session arms the Stop gate for that session (it does not read the conversation; a session without a passing /proof run is not blocked).

Custom bundle & dependency composer

Toggle skills to build a selective install command and inspect required vs. optional branch tools.

npx skills@latest add luci-efe/skills

Required local binaries for selected skills:brew install node git ripgrep ast-grep herdr osv-scanner && uv tool install --python 3.12 mnemosyne-memory
Optional branch tools for selected skills:
  • gitnexus: npm i -g gitnexus (optional · PolyForm Noncommercial license, check before commercial use)
  • playwright: npm i -D @playwright/test && npx playwright install chromium (per-repo local dependency)
  • odiff: npm i -D odiff-bin (optional · per repo (or use Playwright's toHaveScreenshot))
  • cua-driver: from cua.ai (macOS); install not verified by this suite (/doctor only checks that cua-driver is on PATH)
  • gh: brew install gh (Optional read-only Dependabot alert enrichment)

Method · Agents are not chat windows.

01/doctor02/compass03/cartographer04/crucible05/council06/reproduce07/proof08/red-team09/relay01Observable02Not a clipboard03Evidence of done04Context continues05Many models, many harnesses06Autonomy with ownership
  1. Rule 01Observable
    /cartographer /reproduce /proof
  2. Rule 02Not a clipboard
    /council /relay
  3. Rule 03Evidence of done
    /reproduce /proof /red-team
  4. Rule 04Context continues
    /cartographer /relay
  5. Rule 05Many models, many harnesses
    /council
  6. Rule 06Autonomy with ownership
    /doctor /compass /crucible
Text table: Six Rules to skills mapping
Rule/doctor/compass/cartographer/crucible/council/reproduce/proof/red-team/relay
01 Observable○○○
02 Not a clipboard●○
03 Evidence of done●●●
04 Context continues●●
05 Many models, many harnesses○
06 Autonomy with ownership●●●
Fig. 0. Each dot marks a rule enforced by that skill’s decision tree or hook contract. Select any rule or skill to inspect its mechanism.
/compass1-line nudgeMissing tool or hook setupnode/doctor →Unfamiliar domain or call flowgit · rg · ast-grep/cartographer →Unresolved scope or forksrg/crucible →High-stakes design reviewnode · herdr/council →Bug report or regressiongit/reproduce →Advisory or trust exposureosv-scanner · rg · ast-grep/red-team →Claiming work is donenode · input_digest/proof →Context threshold or handoffnode · mnemosyne/relay →
  1. Missing tool or hook setupnode
    /doctor →
  2. Unfamiliar domain or call flowgit · rg · ast-grep
    /cartographer →
  3. Unresolved scope or forksrg
    /crucible →
  4. High-stakes design reviewnode · herdr
    /council →
  5. Bug report or regressiongit
    /reproduce →
  6. Advisory or trust exposureosv-scanner · rg · ast-grep
    /red-team →
  7. Claiming work is donenode · input_digest
    /proof →
  8. Context threshold or handoffnode · mnemosyne
    /relay →
Text equivalent: /compass intent routing and branch prerequisites
  1. Missing tool or hook setupnode/doctor →
  2. Unfamiliar domain or call flowgit · rg · ast-grep/cartographer →
  3. Unresolved scope or forksrg/crucible →
  4. High-stakes design reviewnode · herdr/council →
  5. Bug report or regressiongit/reproduce →
  6. Advisory or trust exposureosv-scanner · rg · ast-grep/red-team →
  7. Claiming work is donenode · input_digest/proof →
  8. Context threshold or handoffnode · mnemosyne/relay →
Fig. 1. /compass routes by intent, probes branch binaries first, and loads only the selected skill tree.

Directory · 09 Skills

  1. 01
    /doctorOrientRule 06

    Setting up, or a skill reports a missing tool, repairing discovery, or configuring the Claude hook adapter.

    Artifact / outputstdout report · .claude/settings.local.json (when wiring hooks)
    Branch toolsnode (required · default)
  2. 02
    /compassOrientRule 06

    Choosing the next engineering skill, discovering installed workflows, or checking prerequisites before a branch.

    Artifact / output.ramos/cache/skills.json
    Branch toolsnode (required · default)
  3. 03
    /cartographerOrientRule 04Rule 01

    You need one domain or flow mapped to real code before planning or editing.

    Artifact / output.ramos/maps/<flow>.md
    Branch toolsgit (required · baseline & gitnexus) · rg (required · baseline) · ast-grep (required · baseline) · gitnexus (optional · PolyForm Noncommercial license, check before commercial use)
  4. 04
    /crucibleDecideRule 06

    A feature, refactor, or design is underspecified or has competing approaches before writing code.

    Artifact / output.ramos/decisions/<topic>.md
    Branch toolsrg (required · default)
  5. 05
    /councilDecideRule 02Rule 05

    A consequential decision or artifact (architecture plan, spec, risky diff, security design) deserves review by several model families in Herdr.

    Artifact / output.ramos/council/<run-id>/SYNTHESIS.md · round1.<seat>.md
    Branch toolsnode (required · default) · herdr (required · default)
  6. 06
    /reproduceBuildRule 03Rule 01

    Something is broken — a bug report, a failing test, a regression, or "it used to work" — before changing any code to fix it.

    Artifact / outputRegression test in repo suite · .ramos/reproduce/{predicate.sh,bisect.log} (bisect)
    Branch toolsgit (required · default & bisect)
  7. 07
    /proofProveRule 03Rule 01

    You are about to say "done" — validating a change, measuring a performance claim, or verifying a web or desktop UI.

    Artifact / output.ramos/runs/<run-id>/result.json
    Branch toolsnode (required · all branches) · playwright (optional · visual-web) · odiff (optional · visual-web pixel diff, or Playwright toHaveScreenshot) · cua-driver (optional · visual-desktop · from cua.ai, install not verified by this suite)
  8. 08
    /red-teamProveRule 03

    Dependencies or lockfiles changed, an advisory or CVE landed, or before a release.

    Artifact / output.ramos/red-team/report.md · .ramos/red-team/osv.json
    Branch toolsosv-scanner (required · scan & github-enrich) · rg (required · scan) · ast-grep (required · scan) · gh (optional · github-enrich)
  9. 09
    /relayContinueRule 04Rule 02

    A session must hand work to a fresh session, another agent, or another machine, or when starting work a previous session handed off.

    Artifact / output.ramos/runs/<run-id>/<pane>/checkpoints/<seq>/handoff.json + HANDOFF.md
    Branch toolsnode (required · default & herdr-launch) · mnemosyne (required · default & herdr-launch · needs uv) · herdr (optional · herdr-launch)

Done vs. verified: once a Claude Code session has a passing /proof run, the Stop gate blocks session exit while evidence is missing, failing, or stale for the current code—unless the run is recorded as blocked, failed, or cancelled with a reason. The local gate prevents accidental false completion claims (a hand-written result can still pass locally); CI re-execution plus independent review earns verified.

Want something like this built for your company?

Book a callMessage me on LinkedInSee Agentic Engineering
← Back to home