Detect plan drift by sampling active work against a plan digest.
npx skills add https://github.com/turbobeest/dev-system --skill plan-guardian
Please help me install this Agent Skill. Skill: plan-guardian Source: https://github.com/turbobeest/dev-system/tree/main/skills/plan-guardian Command: npx skills add https://github.com/turbobeest/dev-system --skill plan-guardian
Background monitoring agent that detects drift from the original plan. It samples work-in-progress and compares against a plan digest to catch misalignment early.
Enable plan guardian and let it run alongside active development to catch drift early.
High-intent search queries and answers about installing and using this skill.
Plan drift detection compares your active development work against a plan digest stored at .claude/plan-digest.json. The Skill samples work streams and phase transitions continuously, raising warnings when misalignment exceeds your configured sensitivity thresholds, so you catch deviations early before they compound.
Yes. Plan guardian runs as a background monitoring agent alongside active development, sampling work-in-progress without blocking agent execution. It injects drift warnings or signals directly into the active agent context when deviations are detected.
Create a lightweight plan digest at .claude/plan-digest.json, enable the plan-guardian-hook.sh sampling script, and configure sampling_rate and sensitivity parameters. Multi-criteria drift assessment then runs automatically against your specified thresholds.
Plan guardian emits structured WARN or DRIFT injections into the active agent context, alongside drift-related signals for audit and decision-making. This allows agents to adjust course or raise alerts based on detected misalignment severity.
Sampling rate controls how frequently work streams are checked against the plan digest, while sensitivity determines the threshold at which deviations trigger warnings. Tuning both lets you balance detection coverage with alert noise.
nnexaiManage multi-phase project milestones via a terminal dashboard.
VSF-QC-TTSAppend structured learning entries to AGENTS_LEARNING.md after AI agent mistakes.
DTMC-marketplaceImplement structured logging and audit trails for EU AI Act compliance.
ajsai47Analyze git diffs for SQL injection, race conditions, and shell injection risks.
