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 skills add MauricioPerera/KDD --skill pm-glm-ccddgit clone --depth 1 https://github.com/MauricioPerera/KDDWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/mauricioperera/kdd/pm-glm-ccdd)<a href="https://agentmods.dev/skills/mauricioperera/kdd/pm-glm-ccdd"><img src="https://agentmods.dev/badge/skills/mauricioperera/kdd/pm-glm-ccdd/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/mauricioperera/kdd/pm-glm-ccdd"><img src="https://agentmods.dev/badge/skills/mauricioperera/kdd/pm-glm-ccdd.svg" alt="Reviewed on agentmods" width="80" height="20"></a>- NVIDIA SkillSpector warn
SkillSpector: 1 finding, up to medium
These are SkillSpector’s own severities. On a checked sample its high-severity flags on skills were ~96% false positives — a documented command, a public API, a “never do X” rule — so we show them as a caution to read, not a verdict. Why →
- medium Agent Snooping · line 128 Skill enumerates or reads other installed skills. Access to other skills' SKILL.md files or the skills directory reveals prompt instructions, capabilities, and secrets that should be invisible to peer skills.Fix: Remove all code or instructions that list or read other skills' files or directories. Skills should operate independently; cross-skill access is a privilege escalation.
What 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 5.1 | $0.00120 | $0.08367 |
| Opus 5 | $0.00060 | $0.04183 |
| Sonnet 5 | $0.00024 | $0.01673 |
| Haiku 4.5 | $0.00012 | $0.00837 |
Grade A, and why
pm-glm-ccdd 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 11d 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 — 346 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Claude = PM · Desarrolladores = instancias efímeras de GLM-5.2 · QA = CCDD gate
Tu rol (PM / orquestador)
- NO escribís código de producción. Tu trabajo: entender el objetivo, descomponerlo en tareas acotadas, redactar specs claras, repartirlas a "desarrolladores" (instancias efímeras de glm-5.2), verificar lo que entregan, integrarlo y reportar estado al usuario.
- Los devs son efímeros: cada tarea = un GLM headless nuevo, SIN memoria entre tareas. Por eso cada spec debe ser AUTOCONTENIDA (objetivo, paths, contexto, restricciones, definición de "hecho").
- La calidad la certifica el CCDD gate (determinista), no tu lectura del código. Leés el veredicto/resumen (PASS/FAIL), no el diff completo. Eso conserva tu contexto (el cuello de botella).
Tiering de modelos (elegí el tier por tarea; no gastes el caro donde no aporta)
Pipeline por defecto, del más barato al más caro. Cada tier es un Agent con model: distinto
(o el propio main-loop). El gate CCDD nivela el downside de usar tiers baratos: sus errores afloran
como tests rojos / violaciones de budget.
- Triage → Haiku 4.5 (
model: "haiku"). Tareas de clasificación/extracción deterministas al arrancar: leer un issue, clasificar el tipo de bug/subsistema, limpiar/resumir un log de errores, mapear las tools MCP necesarias, y detectar si el trabajo ya está hecho. Barato y rápido. Verificado (2026-07-05): un triage Haiku (~56k tokens) detectó que un issue YA estaba arreglado en HEAD → cortó el pipeline antes de disparar PM+dev sobre un bug inexistente. Ese es su mayor valor: evitar gasto caro. REGLA: verificás su reporte por artefacto (código/commits/suite) antes de actuar — es tier barato, puede errar; el gate/tu verificación es la ley. - PM/orquestador → Sonnet 5 (
model: "sonnet") POR DEFECTO. Autora contrato + tests congelados, delega a GLM, verifica por gate. Verificado (2026-07-05) en un A/B medido: Sonnet 5 diseñó un oráculo MÁS riguroso que Opus (cazó 4/4 bugs plantados vs 2/4) por ~97k tokens — igual-o-mejor calidad a ~1/5 del costo en tareas de patrón conocido (validación, CRUD, wrappers, parsing). - Dev/implementador → GLM (
glm-5.2:cloudvíaollama launch claude). El código de producción. - Reserva → Opus 4.8 (main-loop o
model: "opus"). SOLO tareas genuinamente novedosas / alto riesgo donde el discriminador adversarial NO es un tropo conocido (p.ej. inyectar crash enfs.renameSync, anti-degradación de over-fetch que exige insight del modo de fallo), + juicio e integración final.
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.
- 11d ago First seen · 346 lines · 120 tokens per session scan A bbcb1c1381bb
pm-glm-ccdd is a skill published in the GitHub repository MauricioPerera/KDD (7 stars, last pushed 2d ago), licensed MIT. It adds 120 tokens to every session and 8,367 once invoked, about $0.0006 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 skills, from other repositories
crit-cli
Use when an agent needs to author or reply to crit inline comments programmatically (including multi-agent workflows commenting on shared code/plans/docs/proposals), publish or unpublish a crit review with crit share, sync a crit review to or from a GitHub PR or GitLab MR, or read/interpret a crit review JSON file.…
crit
Review code changes, a plan, a live page (running dev server), or a local HTML file with Crit inline comments and structured human feedback. Use only when the user explicitly invokes /crit or directly asks to use Crit; a generic review request does not count.
crit-story
Author a crit story and continue the interactive review loop only when the user explicitly invokes crit-story or directly asks you to generate a crit story. Do not infer this skill from generic review, PR, or diff-review requests.
sdlc-accelerate
End-to-end SDLC ramp-up from idea to construction-ready with automated phase transitions.
workspace-prune-working
Clean up .aiwg/working/ by promoting, archiving, or deleting temporary files.
workspace-realign
Reorganize and update .aiwg/ documentation to reflect current project reality.