Investigate bug reports or feature requests and document findings as a structured .planning dossier.
npx skills add https://github.com/bravros/bravros --skill recon-bravros
Please help me install this Agent Skill. Skill: recon Source: https://github.com/bravros/bravros/tree/main/plugins/core/skills/recon Command: npx skills add https://github.com/bravros/bravros --skill recon-bravros
.planning/P-NNNN-<slug>/ folder with evidence, per-issue files, decisions, traps, and acceptance criteria, each claim tagged with a confidence level (CERTIFIED, OBSERVED, READ, ASSUMED, SUPERSEDED). - Evidence discipline: Copies attachments into an evidence/ folder, records what each artefact shows rather than conclusions, and requires a falsifier statement on every issue to guard against confident wrong premises. - Clean handoff to orchestration: Documents what is wrong and what is wanted without prescribing execution order, then hands off to /orchestrate with a named branch and a single next step. - Use Case: A user reports that a dashboard chart rescales on scroll. Invoke /recon chart rescales on scroll to classify it as a defect, gather runtime evidence, and produce a reviewed dossier ready for planning. ## Quick Start Ask the agent to run /recon on your bug report or feature idea, for example by saying: investigate why the miner page chart rescales on wheel scroll and document it as a planning dossier.High-intent search queries and answers about installing and using this skill.
Run /recon with a description of the problem. It classifies the issue as a defect or change, gathers ground truth from code and runtime evidence, and writes a .planning/P-NNNN dossier with per-issue files, confidence tags, and acceptance criteria ready for planning.
Recon produces findings only: what is wrong, what is wanted, and what is true, with no phases or ordering. Orchestrate consumes that dossier and makes all execution decisions such as sequencing, grouping, and parallelism.
Yes, invoking /recon with the --worktree flag creates an isolated worktree and branch using the bravros worktree setup command, and writes the dossier there so the main checkout stays untouched for other running sessions.
Confidence tags (CERTIFIED, OBSERVED, READ, ASSUMED, SUPERSEDED) distinguish runtime proof from code-reading hypotheses. This prevents a wrong premise stated as fact from being built on, which the documentation identifies as the most expensive recon failure.
An UNCERTIFIED outcome is valid. The dossier then documents the next investigation steps rather than a fix, and claims are tagged ASSUMED with the command or question that would settle them.
SwitchbackTechConvert user bug reports into detailed GitHub issues with evidence and diagnosis.
robdmcExecutes implementation plans by coordinating parallel task agents and dependencies across phases.
PeriscopeAIDiagnose workflow execution failures and provide optimization recommendations.
enigmaicon-engStructure multi-agent debates into synthesized decisions with documented disagreements.