import 96/100 — SaferSkills

CategoryWeightCategory scoreContribution

Securityprompt, exec, net, exfil, eval

35%

30.8 pts

Supply chainhash, typosquat, maintainer, lockfile

20%

100

20.0 pts

Maintenancestaleness, pinning, CI

15%

100

15.0 pts

TransparencySKILL.md, perms, README

15%

100

15.0 pts

Communityinstalls, verify, response

15%

100

15.0 pts

Findings & checks · 1 flagged

Securityscore 88 · 1 finding

MEDIUMInstruction telling the agent not to ask for approvalSS-SKILL-INJECT-DONT-ASK-01 · Prompt injection · CHANGELOG.md

MEDIUM — it fires on intent; the real damage depends on the host agent's own approval-gating.

Why it matters

The text {match} tells the agent to skip the normal "ask the user first" gate. Used adversarially it removes the human-in-the-loop check before destructive or sensitive actions, turning a normally-gated agent into a fire-and-forget executor.

The exact value spotted

excerptCHANGELOG.md· markdown

551- Setup Step 10 checks whether the Claude Code task-list tools (TaskCreate/TaskList) are opt

… (107 chars elided on L551)

552- New helper `scripts/credo-todo-tools.sh` (status, enable, decline, undecline, periodic hin

… (98 chars elided on L552)

553- New SessionStart hook `credo-todo-tools-hint.sh`: while the opt-in is off, it offers it ag

… (108 chars elided on L553)

554- The recommendation also mentions that Ctrl+T shows or hides the task list, since the list

… (49 chars elided on L554)

555

Occurrences

1 occurrence · at L553

How to fix

Remove the approval-skipping instruction, or scope it narrowly to a specific safe, reversible action.

Delete blanket "don't ask / no need to confirm" directives from the skill.
If the skill is a genuine autonomous job, restrict the opt-out to a named non-destructive action rather than all actions.

Framework references

OWASPLLM01 ATLASAML.T0051

Trace & refs

ruleSS-SKILL-INJECT-DONT-ASK-01sha25614b7852251bf4666…rubric 365aacaView on GitHub

Supply chainscore 100 · 0 findings

✓—All supply chain checks passedNo findings in this category for the latest scan.pass

Maintenancescore 100 · 0 findings

✓—All maintenance checks passedNo findings in this category for the latest scan.pass

Transparencyscore 100 · 0 findings

✓—All transparency checks passedNo findings in this category for the latest scan.pass

Communityscore 100 · 0 findings

✓—All community checks passedNo findings in this category for the latest scan.pass

The primary manifest — the file an agent reads to learn what this artifact does.

README.md · 1 file

README.md13.3 KB · Markdown

<h1 align="center">marcel-bich-claude-marketplace</h1>

Your AI got promoted. Now it needs management.

Claude Code isn't just a tool. It's your personal employee. A brilliant, tireless, and - more often than not - occasionally chaotic employee.

Compliance Officer. Accountant. Project Manager. Receptionist. Mentor. Librarian. Taskmaster. These plugins handle most of that - so you can focus on building.

Why?

You asked Claude to commit your changes. He force-pushed to main. On Friday. At 5pm.

You ran out of API tokens mid-task. No warning. Context gone.

Main agent working hard but starts degrading after 50% context window? Hooks with subagents and Hydra with parallel work save your day.

You needed parallel work. Git conflicts everywhere.

You left Claude alone for 5 minutes. He's been "thinking" ever since - and you only know if you check yourself (Schroedinger's Claude).

Claude is brilliant. Claude is also a golden retriever with a keyboard - enthusiastic, helpful, and completely unsupervised with root access.

These plugins add the leash.

(Yes, I know full well you're all bypassing permissions because it's FAST. But you should at least put in SOME form of guardrails and check in occasionally to guarantee quality - not just blindly let it run and hope for the best, only to end up with buggy or even insecure AI slop.)

Without guardrails:

Context dies mid-task? Start over.
API limit hit? Surprise!
Need parallel work? Git conflicts incoming.
Was that commit good? Who knows, Claude already pushed.

The Solution

A growing collection of plugins, continuously developed to meet my extremely high expectations for AI-assisted development.

Born from real pain points. Built for developers who learned the hard way - since Claude Code exists.

Each plugin works independently - pick what you need. But together, they form a solid framework for developers who want control, quality, and security over pure YOLO mode.

credo is the mentor - it governs how a whole session runs, and the other tools slot in around it.

They don't replace Claude. They make Claude manageable.

Plugin	Role	What they do
credo	The Mentor	"Here is how we run a session." Modes, work-items with a hard Definition of Done, budgeted autonomy, verify, and safety everywhere.
dogma	The Compliance Officer	Intelligent rule sync from any source, with enforcement hooks. Opinionated defaults included.
hydra	The Project Manager	Parallel workstreams. Isolated branches. No stepping on toes.
signal	The Receptionist	"Claude needs you." Ding. Simple.
limit	The Accountant	Tracks every token. Shows the burn rate and limits. No surprises.
import	The Librarian	External docs, locally cached. No annoying "site blocked AI."
marketplace	The Intern	Development tools for local plugin testing.

Third-Party Plugins

Plugin	Author	Role	What they do
get-shit-done	Marcel-Bich (secured fork)	The Taskmaster	Specs first. Implementation second. Ship third. An optional alternative to credo's item workflow for spec-driven planning.

credo - the mentor

credo is the mentor - a self-contained process framework that governs how a whole session runs. Where the other plugins each solve one problem, credo sets the working discipline for the whole session and the rest slot in around it:

Session modes - active, passive, and autonomous. Each mode changes how much Claude does on its own and whether it keeps itself alive for unattended work. The current mode is re-injected on every prompt, so it is never silently forgotten.
Work-item lifecycle with a hard Definition of Done - items live under .credo/, where the folder is the status (clarify -> go -> done -> verified). Nothing reaches "done" until an independent audit subagent signs off and, for UI work, a real browser verify captured screenshot evidence.
Budget and limit awareness - credo reads the 5-hour and weekly usage caps, sizes tasks to fit, and pauses or hands off before a wall is hit.
Verify and wiring checks - "done" means the code is actually wired in and observably works, not just "the test passed".
Safety that travels - filesystem protection and no-autonomous-installs rules are re-injected into every subagent, so delegation cannot dilute them.
Autonomy handling - approved GO items are worked unattended with hook-enforced self-scheduled wake-ups (a registered Stop hook blocks a stop without a scheduled ScheduleWakeup), budget caps, and ntfy notifications per task and question.
Subagent priming - every subagent starts with the load-bearing security, quality, and honesty rules already in context.
WSL environment awareness - reach Windows-side services and launchers from WSL, self-detecting.
Opt-in versioning - keep .credo/ local by default, or version items and process in the repo with a single flag.

Start here: install credo, run /credo:setup, then pick a session mode. The rest of the marketplace slots in around it.

credo vs Get-Shit-Done

credo has two layers, and only one of them is a task system:

Operating layer (always on) - session modes, WSL environment awareness, budget and limit awareness, subagent orchestration and priming, safety that travels into every subagent, and the session-mode switch. This layer is orthogonal: it sits on top of whatever task system you use.
Task model (.credo/items/ plus a hard Definition of Done) - this is an ALTERNATIVE to GSD, not a layer on top of it. Use one task system per project.

credo is the recommended default and is self-contained for day-to-day work - items, bug fixes, incremental features, each gated by an observable Definition of Done. credo deliberately has NO large up-front project bootstrap or roadmap; it stays flat and reactive.

GSD is optional for heavy up-front planning (PROJECT -> ROADMAP -> milestones -> phases -> plans), or if you already know and prefer the original GSD flow. Pick ONE task system per project, never both at once.

If you run GSD as your task system, set the task backend to gsd so credo's own item features stand down - that avoids two competing sources of truth (.credo/ items vs GSD's .planning/). The backend is config-driven: it lives in .credo/config as task_backend (default credo), and /credo:setup writes gsd there for you when you pick GSD. The CREDO_TASK_BACKEND environment variable is an override if you ever need it. Any value other than gsd keeps credo's items active; the operating layer above stays on regardless of the value.

Technically the two do not collide: their hooks are disjoint, and their commands and directories are separate. The one point of friction is the status line - it is a single slot - so in a combined credo + GSD + limit setup keep the limit status line (let GSD skip installing its own) so credo's budget and auto-compact features keep their data source.

Requirements

All plugins require:

jq - JSON processor (install: sudo apt install jq / brew install jq)
Supported: Linux, macOS, WSL2
Not supported: Native Windows / PowerShell (use WSL2 instead)

Strongly recommended: Run Claude Code inside tmux (also inside WSL2). credo's /credo:self-compact and /credo:self-reload only work in a tmux pane, and /credo:self-restart only waits for an idle input field there. /credo:setup detects tmux, shows the install command for your package manager and runs it only after your yes. Details: credo wiki - tmux.

Quick Start

claude plugin marketplace add Marcel-Bich/marcel-bich-claude-marketplace
claude plugin install credo@marcel-bich-claude-marketplace
claude plugin install dogma@marcel-bich-claude-marketplace
claude plugin install hydra@marcel-bich-claude-marketplace
claude plugin install signal@marcel-bich-claude-marketplace
claude plugin install limit@marcel-bich-claude-marketplace
claude plugin install import@marcel-bich-claude-marketplace
claude plugin install get-shit-done@marcel-bich-claude-marketplace

Documentation

Full documentation, installation, configuration, and troubleshooting:

[View Documentation on Wiki](https://github.com/Marcel-Bich/marcel-bich-claude-marketplace/wiki)

License

MIT - See LICENSE for full terms.

Contributing

I'm fine with issues containing code snippets, and I'm not against pull requests, but please read CONTRIBUTING.md before submitting anything! The contributing rules only exist to prevent licensing issues on my end.

Name, Logo, Trademarks (No endorsement)

The source code is licensed under MIT. However, the project name "marcel-bich-claude-marketplace", logos, and branding assets are not covered by the MIT License.

Trademark rights are not granted by the MIT License. Using the project name or branding in ways that suggest endorsement, official affiliation, or sponsorship is not permitted.

Forks and derivative works must use a different name and their own branding. A clear statement that your project is not official and not affiliated is required.

See TRADEMARK.md for the full trademark policy.

Star History

<details> <summary>Keywords / Tags</summary>

Claude Code, Claude Code Plugin, Claude Code Extension, Claude Code Marketplace, Claude Code Plugins Collection, Anthropic CLI, Anthropic Plugin, Anthropic Extension, Anthropic Claude, Anthropic AI, AI Agent Plugins, AI Code Assistant, AI Coding, AI Programming, AI Development, statusline, Statusline, Status Bar, API Usage, Desktop Notifications, Git Worktree, Parallel Agents, Documentation Import, Context7, Playwright, Rules Enforcement, Workflow Guides, Best Practices, Marcel Bich, marcel-bich-claude-marketplace, dogma plugin, signal plugin, limit plugin, hydra plugin, import plugin, credo plugin, get-shit-done plugin

</details>

The preacher invites the curious to study the [Genesis](GENESIS.md).

import— plugin

Is import safe to install?

Why?

The Solution

Third-Party Plugins

credo - the mentor

credo vs Get-Shit-Done

Requirements

Quick Start

Documentation

License

Contributing

Name, Logo, Trademarks (No endorsement)

Star History

Audit the pieces. Scan the whole. Decide.