Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add commands/vsladkov/claudex-stereo/implementgit clone --depth 1 https://github.com/vsladkov/claudex-stereoWhat it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.00015 | $0.15037 |
| Opus 5 | $0.00008 | $0.07519 |
| Sonnet 5 | $0.00003 | $0.03007 |
| Haiku 4.5 | $0.00002 | $0.01504 |
Grade A, and why
implement scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 2d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 1,081 lines — stays where its author put it; the contents beside it link to each section on GitHub.
First Read ${CLAUDE_PLUGIN_ROOT}/skills/model-routing/SKILL.md and apply its routing, foreground
agent, validation, quoting, and background-job rules. The rules below are step-specific.
Run the implementation phase of the Stereo workflow. The main Claude session owns preflight, baselines, host verification, review/fix orchestration, and the final report.
Raw slash-command arguments:
$ARGUMENTS
Arguments and modes
After reading the routing skill, parse all arguments before loading state:
--implementer <model>selects the implementer. Resolve it as explicit flag > workspaceimplementerdefault >claude:opus.--implementer-effort <none|minimal|low|medium|high|xhigh|max>overrides effort for a Codex-routed implementer.--implementation-reviewer <model>selects the implementation reviewer. Resolve it as explicit flag > workspaceimplementationReviewerdefault >codex:sol: the implementation review is the last gate before commit, and the cross-ecosystem reviewer is independent of both the orchestrator and the Claude-routed default implementer that produced the delta.claude:sessionremains valid and is the cheaper inline choice.--implementation-reviewer-effort <none|minimal|low|medium|high|xhigh|max>overrides effort for a Codex-routed implementation reviewer.--effort <none|minimal|low|medium|high|xhigh|max>is the command-wide default for Codex-routed roles that have no role effort flag. When no active role is Codex-routed, a command-wide--effortis inert: accept it, report it as inert, and never translate it into a Claude-side control.--max-fix-rounds <n>defaults to 4. Direct gate-fix turns and review-driven fix turns both count toward it; the implementer's own inner-loop iterations never do.--slot <name>selects the stored plan to implement and defaults todefault.--freshskips a reusable stored Codex plan-review thread; it only affects a Codex-routed implementer and is accepted but reported as inert for a Claude-routed one.--implement-onlyimplements and verifies once, then stops before review.--review-onlyreviews the current dirty/untracked implementation delta once without fixing.--resumere-enters a recorded incomplete implementation phase after a crashed, compacted, or closed Claude session.--isolatedruns implementation in a throwaway detached git worktree outside the repository and hands the delta back as a user-confirmed patch. It never commits and never writes in the main working tree.--base <ref>reviews the committed<ref>...HEADrange and is valid only with--review-only.- Without a mode flag, run the complete implement-plus-review/fix phase.
Define <slotArg> as --slot <slot> for a non-default target slot and omit it entirely for the
default slot. On --resume, replace it with the slot owned by the durable implementation record.
Reject missing values, duplicates, positionals, invalid effort/round values, unknown flags,
unknown claude:* values, and both mode flags together. Accept claude:inherit alongside
claude:session and the four explicit Claude aliases. Accept a Codex selection with or without
the codex: prefix and reject codex:claude:*. The removed implementer --model and --review-model
flags are unknown; report the role-named replacements.
The renamed --impl-reviewer and --impl-reviewer-effort flags are unknown; report
--implementation-reviewer and --implementation-reviewer-effort as their replacements.
For --review-only, reject --implementer, --implementer-effort, --fresh, and
--max-fix-rounds. For --implement-only, reject --implementation-reviewer,
--implementation-reviewer-effort, and --max-fix-rounds. Reject a role effort flag when its
selected role is Claude-routed.
Reject --resume with --implement-only, --review-only, or --fresh. Reject --resume with
--implementer or --implementer-effort: the durable record owns the implementer; tell the user
to run without --resume to start over. Reject --resume with --slot: the durable record owns
the plan slot; tell the user to run without --resume to start over. --implementation-reviewer,
--implementation-reviewer-effort, --effort, and --max-fix-rounds remain legal with
--resume. Reject --base without --review-only. Reject --scope in every mode and name
--base <ref> as the supported standalone range control: auto-detecting a default base could
silently review commits the plan never covered.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 2d ago First seen · 1,081 lines · 15 tokens per session scan A fa906acf58e6
implement is a command published in the GitHub repository vsladkov/claudex-stereo (3 stars, last pushed 23d ago), licensed Apache-2.0. It adds 15 tokens to every session and 15,037 once invoked, about $0.0001 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other commands, from other repositories
audit-agents-skills
Audit quality of agents, skills, and commands in a Claude Code project.
audit-repo-docs
Audit repository documentation against 85+ best practices from claude-code-ultimate-guide.
boldguy-linkedin-answer
Draft a LinkedIn comment on a post, referencing the guide's relevant section, in Flow style with AI markers removed.
security-audit
Comprehensive security audit of your project AND Claude Code configuration. Analyzes secrets exposure, injection surfaces, dependencies, hook security, and produces a scored security posture assessment.
audit-quiz-coverage
Find quiz coverage gaps from recent guide/CHANGELOG/CC-releases changes and propose new questions.
retex
Retex - Capture lesson learned dans memory après fix, rollback, erreur.