Finding
Small Hermes tasks become operationally expensive when the agent keeps expanding scope instead of stopping early to re-scope, batch, or defer the new work.
Current
A real Hermes installation often starts with a tiny request: patch one prompt, update one page, fix one config detail, check one route, or adjust one skill. The weak point appears when each discovery creates another “quick” follow-up, and the session turns into an expedition across files, tools, docs, cron jobs, skills, memory, and public copy. Without a re-scope checkpoint, the agent can spend tokens and attention chasing adjacent improvements that were never part of the original outcome.
Suggested
- Add a scope-tripwire for “quick” tasks. Exact change: add a short rule to
SOUL.mdor the main operator instructions: “If a quick task reveals more than two adjacent fixes, stop and report the original task status, the new findings, and a recommended batch plan before continuing.” - Create a Vibe Coding re-scope checklist. Exact change: add
docs/runbooks/scope-control.mdwith this checklist: define the original ask, list discovered extras, classify each as required/blocking/optional, complete only blocking items now, and move optional items to a follow-up batch or Kanban card. - Add a verification habit before expanding work. Exact change: patch the relevant coding or content skill with: “Before making an unrelated file edit, prompt change, cron adjustment, or dashboard copy update, verify whether it is necessary for the current acceptance criteria or should be captured as a follow-up.”
Impact
This keeps Hermes work aligned with the user’s actual intent instead of letting discovery drive unlimited execution. It improves prioritization because necessary fixes still happen, while optional improvements are grouped into visible batches. It also reduces token waste and operational risk by avoiding broad edits that are hard to review, test, or explain afterward.
Effort
Small — the fix is mostly one profile rule, one lightweight runbook, and one recurring verification habit. No new infrastructure is required; the main discipline is pausing early when “quick” stops being true.
Public page note
Safe public content includes the achievement theme, the scope-control principle, generic examples of task expansion, and the recommendation to re-scope early. Internal-only content includes private task details, raw session excerpts, live file paths, logs, credentials, config values, customer-specific workflows, and any unresolved operational findings from a real installation.