Logs local and free model failures to a markdown file for future Grok escalation decisions.
npx skills add https://github.com/GreenyZA/neo-light --skill bring-grok-greenyza
Please help me install this Agent Skill. Skill: bring-grok Source: https://github.com/GreenyZA/neo-light/tree/main/.grok-home/skills/bring-grok Command: npx skills add https://github.com/GreenyZA/neo-light --skill bring-grok-greenyza
when-grok.md file whenever a turn fails on local or free tiers. - Escalation Evidence: Builds a real-world brief over time that shows exactly which task categories justify premium model spend. - Use Case: A coding task fails on a 3B local model due to context limits. The Skill logs the failure row, and after enough rows accumulate, the team has concrete data to write the Grok usage playbook. ## Quick Start Tell the agent to log this failed turn with bring-grok, noting what failed and which lane was tried.High-intent search queries and answers about installing and using this skill.
Log each failed turn as a row in a markdown table with the date, task, failure reason, and lane tried. This Skill appends that row to a when-grok.md file whenever local or free models cannot finish a task.
Escalate only after real failure rows accumulate showing repeated quality, context, or tool-calling failures on local and free lanes. This Skill builds that evidence file instead of guessing when premium spend is worth it.
No. The Skill explicitly does not call SuperGrok from this environment. It only records failure evidence so a human can later decide when bringing in Grok is justified.
Each row records the date, the task attempted, what failed, which lane was tried (vulkan, ollama, minimax, or openrouter-free), and why Grok would help. No cost estimates or full prompt dumps are stored.
Markdown logging requires manual review to spot patterns and does not automate escalation decisions. It suits small teams building evidence before committing to paid model routing rules.
nabitllcRoutes agent turns across three local model tiers and flags turns exceeding twice the median duration.
BoardAIRoute contract issues to the correct approver and draft the escalation message.
ruvnetAnalyze token usage patterns and recommend model and caching optimizations with estimated savings.
Arete-ConsortiumDiagnose CI build failures and propose minimal fixes from GitHub Actions and GitLab logs.