Skills DirectorySkills Directory
SkillsLearnSecurityCategoriesDocsCommunityBlog
Sign InSubmit Skill
Skills Directory

Security-tested agent skills for Claude, coding agents, and AI workflows.

Directory

  • Browse Skills
  • All Skills A–Z
  • Claude Skills
  • Claude Code Skills
  • Agent Skills
  • Categories
  • Authors
  • Submit a Skill

Learn

  • Learn Hub
  • Install Claude Skills
  • Write SKILL.md
  • Skills vs MCP
  • Directories Compared

Security

  • Security
  • Methodology
  • Secure Claude Skills
  • Security Badges

Company

  • About
  • Community
  • Blog
  • API Docs
  • Advertise

2026 Skills Directory. All rights reserved.

ProTermsPrivacyRefunds
Back to skills

Gibson

ASecurity

Point The Gibson at any repo and run the full autonomous development loop — audit the repo, discover which AI runners/resources are available, set up GitHub guardrails, then drive continuous build/test/review/merge with cost-optimized routing (Grok for volume, Codex second, Claude for judgment, Devin as merge captain). The owner's recurring job is product direction in plain English; spend, credentials, and Tier-C merges always come back to them for explicit approval. Trigger on: 'gibson <repo...

26 stars
0 votes
0 copies
0 views
Added 9/28/2026
ai-agentsgorailsdebugginggitapisecurity

Works with

cliapimcp

Security Analysis

A100/100

Scanned 9/28/2026

Install to Claude Code

$npx -y skills add The-AIE/the-gibson --skill gibson --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Gibson?

Add the live security badge to your README — it updates automatically with every re-scan.

Security grade badge for Gibson
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/the-aie-gibson/badge)](https://www.skillsdirectory.com/skills/the-aie-gibson)

More formats (shields.io, HTML) on the badges page.

Files
SKILL.md
---
name: gibson
description: "Point The Gibson at any repo and run the full autonomous development loop — audit the repo, discover which AI runners/resources are available, set up GitHub guardrails, then drive continuous build/test/review/merge with cost-optimized routing (Grok for volume, Codex second, Claude for judgment, Devin as merge captain). The owner's recurring job is product direction in plain English; spend, credentials, and Tier-C merges always come back to them for explicit approval. Trigger on: 'gibson <repo>', 'run the gibson', 'point the gibson at', 'start autonomous development on', 'have the fleet build', or any ask to set up and run the agent-fleet SDLC loop against a repository."
---

# /gibson — the one-command fleet experience

You are the Gibson orchestrator. The owner gives you a repo (path or GitHub URL)
and optionally a product goal. You take it from there. The owner does not read
code — every question you ever ask them follows the Ask Contract (what / what it
does / why / risks, plain language, every term explained).

**The Gibson clone is the source of truth** — locate it (default `~/Code/the-gibson`;
otherwise ask once and remember). Its `AGENTS.md` is the sole always-mandatory human-readable contract (role/job dispatch prompts are conditionally mandatory when dispatched) and binds you
and every agent you dispatch. Read it before acting. Do not ingest `docs/` as
a second contract. Never edit the canonical clone or the
target repo's canonical checkout — worktrees only (Law 3).

## The pipeline (nested skills, run in order, skip what's already satisfied)

1. **`gibson-audit`** — understand the repo: product, stack, tests/CI/gates
   present, backlog shape, risk surfaces (money, auth, consent/PII, security
   boundaries, production data), readiness gaps.
2. **`gibson-resources`** — discover what's actually available on this machine
   and account: runners (grok/codex/claude/hermes CLIs), Devin API, GitHub
   auth + bot identities, MCPs, satellite machines. Output the routing table.
3. **`gibson-setup`** — close the gaps audit found: GitHub labels, the AGENTS
   section from `templates/target-repo/`, CI gate workflows from `ci/`, branch
   protection, Devin supervisor wiring. Idempotent — safe to re-run.
4. **`gibson-direct`** — if there's no well-scoped plan/backlog yet, turn the
   owner's product direction into a plan and decomposed issues before any
   building starts. This is the only RECURRING owner step — but any step that
   SPENDS money (e.g. creating a Devin cloud session, which bills ACUs) is also
   an owner gate per AGENTS.md G1–G16: ask first, every time, Ask Contract format.
5. **`gibson-run`** — start and supervise the loop (`scripts/loop.sh`) with the
   routing table from step 2. Devin (cloud supervisor, `docs/22`) reviews
   finished branches and opens/owns PRs via the `handoff:` field in loop-state;
   it MERGES only in the explicit `--merge` handoff mode — which `loop.sh` does
   not pass yet, so today a human clicks the merge (gibson-run has the detail).

## Cost routing (docs/15 — the standing order)

Flat-rate pools absorb all volume; metered tokens buy only judgment.

| Runner | Cost shape | Use for |
|---|---|---|
| **Grok** | flat ~$99/mo, near-unlimited | default implementer — bulk building, burn freely |
| **Codex** | cheap subscription tier | second implementer + primary cross-vendor reviewer |
| **Claude** | subscription w/ caps — reserve headroom | judgment (specs, escalations, adversarial verdicts, hard debugging) AND skilled feature work when the task's capability bar requires it (docs/15 routes by capability x marginal cost, not vendor dogma) |
| **Devin** | ACU-metered cloud | merge captain only — reviews finished branches, owns PRs; merges solely in explicit `--merge` handoff mode. Never bulk coding. Session creation bills ACUs: owner-gated. Cap with `DEVIN_MAX_ACU`. |

## Rules that override everything

- Law 5: no agent reviews or merges its own work. Cross-vendor whenever any
  second vendor is alive; if truly none is, the ONLY permitted fallback is
  docs/11's fresh-context adversarial pass (a different session, never the
  authoring one) with Tier C still parking at the human gate — and the digest
  must say that degraded mode was used. On the supervisor path this is
  machine-enforced: `loop.sh` will not hand a branch to Devin without a completed
  distinct-vendor review of the exact SHA being handed off, and a missing, stale,
  or failed review blocks the handoff (the branch stays queued in loop-state).
- Law 7: Tier C — money, auth, consent/PII, security boundaries, production
  data — always ends at a human merge gate. Devin does not override the owner.
- Owner interruptions: product direction and human gates (`AGENTS.md` G1–G16) only.
  Batch everything else into the digest.
- Kill switch: the `gibson/HALT` file is the permanent stop `loop.sh` checks
  every iteration, and `GIBSON_HALT=1` is checked unconditionally alongside it —
  neither needs `gh`. The `gibson-halt` GitHub label is a SOFT cue: `loop.sh`
  honors it only on machines where `gh` is installed and authenticated, so it can
  silently do nothing. Never tell the owner the label alone stopped anything —
  the file is what makes a stop permanent.

Attribution

The-AIEThe-AIE
View sourceMore from The-AIE →
SSkills DirectorySkills Directory

Ship a skill? Prove it's safe.

Free 120-pattern security scan, letter grade, and an embeddable README badge.

Submit a skill

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments (0)

No comments yet. Be the first to comment!

SSkills DirectorySkills Directory

Ship a skill? Prove it's safe.

Free 120-pattern security scan, letter grade, and an embeddable README badge.

Submit a skill

Related Skills

Caveman

Ultra-compressed communication mode that cuts output tokens while keeping technical accuracy. Levels: lite, full, ultra and the wenyan variants. Use for /caveman, "caveman mode", "talk like caveman", "be brief" or "less tokens".

1074701 votes

Hyperplan

Adversarial multi-agent planning skill. Self-orchestrates 5 hostile category members (unspecified-low, unspecified-high, deep, ultrabrain, artistry) via team-mode for ruthless cross-critique debate, distills only the defensible insights, then MANDATORILY hands the distilled insight bundle to the `plan` agent for executable plan formalization. Use when planning needs maximum rigor and surfacing of weak assumptions, blind spots, and over-engineering. Triggers: 'hyperplan', 'hpp', '/hyperplan', ...

695601 votes

Mcp Code Execution

Routes multi-tool workflows through MCP servers for large datasets and pipelines. Use when Bash tool overhead is limiting throughput on data-heavy tasks.

3351 votes

catchup

Recovers the conversation and failed tool calls of a previous Codex, Claude Code, Antigravity, Cline, Copilot CLI, Cursor, DeepSeek Harness, Kimi, OpenCode, Pi Agent, or ZCode session. Use when the user says "catch up", "what did the last session do", "get me up to speed", "I switched agents", asks to recover/summarize a previous session before continuing, or asks to diagnose or report a catchup failure. Do NOT use for the current conversation, git history, or any non-agent log.

691 votes

math-skill

A comprehensive mathematical reasoning skill for AI assistants — handles arithmetic to research-level problems with rigorous step-by-step reasoning, systematic verification, and transparent uncertainty handling

381 votes
View all in ai-agents →