SoloHQ watches how agents work in each portfolio company, fixes what holds them back on its own, and keeps the context layer they run on current.
No surveys, no workshops. SoloHQ connects to each company's repos, docs, and agent logs, and works out on its own where the context layer let an agent down: conflicting knowledge, errors without a cause, checks too slow to use.
SoloHQ writes the change, replays the runs that failed, and ships it when they pass. Stale context gets rewritten, swallowed errors get their cause back, slow checks get a fast path. People see only the calls that need judgment.
The context layer is a plain directory per company: reconciled knowledge and shared skills. SoloHQ keeps it current and rewrites a skill when runs show it has drifted. Agents read it with ls, grep, and cat. MCP and REST are there where they fit better.
Connect its repos, docs, and agent logs, and let the loop build its context layer. Extend it to the rest of the portfolio when you like what it fixes.