Spawns three parallel review subagents over a session transcript to route learnings into skill edits.
npx skills add https://github.com/edivad1999/stuc-stack --skill reflect-edivad1999
Please help me install this Agent Skill. Skill: reflect Source: https://github.com/edivad1999/stuc-stack/tree/main/skills/reflect Command: npx skills add https://github.com/edivad1999/stuc-stack --skill reflect-edivad1999
references/. - Structured synthesis: A synthesizer subagent merges findings into an Accepted / Rejected / Backlog list with durability, specificity, and convergence criteria. - Approval-gated application: Accepted edits are presented to the user before landing; trivial edits are applied directly while substantive ones route to the create-skill draft/test/iterate loop. - Use Case: After a debugging session where the agent discovered a non-obvious Gradle flag and a missed skill trigger, run reflect to capture the flag convention in the relevant skill and tune the missed skill's description so it fires next time. ## Quick Start Say "reflect" after a complex task completes to review the session and route durable learnings into skill edits.High-intent search queries and answers about installing and using this skill.
Invoke the reflect skill after a complex task completes. It locates the active transcript, spawns three parallel reviewer subagents with different lenses, synthesizes their findings, and routes approved learnings into edits on existing skills.
The reflect reviewers detect missed-trigger cases where a visible skill would have helped but was never invoked. These findings route as "tune description" items, which are handed to the create-skill description-optimization loop so future agents pick the skill up.
Run it when the user says "reflect", after a complex task with 5+ tool calls lands cleanly, when the agent hit dead ends before finding a generalizable path, or when the user corrected the approach mid-task. Skip trivial or already-covered sessions.
No. The synthesizer's full Accepted/Rejected/Backlog output is presented to the user first, and edits land only after explicit approval. Backlog items are filed to the team's devex tracker automatically, but skill edits never auto-apply.
Findings must scope to skills, tools, or MCPs actually invoked in the transcript; speculative routings to unopened skills are dropped. Implementation details that drift, such as SHAs, file paths, and version numbers, are rejected in favor of durable principles.
cursorDrafts a personal mode skill capturing user working conventions from transcript mining.
stellaroneIteratively improve Claude skill descriptions through trigger evaluations and structured edit loops.
cursorConverts recurring corrections into lint rules, runtime checks, and automation scripts.
aslaiiExecute trivial tasks directly without spawning subagents or creating PLAN.md files.