proofline-lead

Purpose

The proofline lead's lifecycle method for the main session (AGENT_ROLE unset = lead).

Source Skill purpose

When to use it

Use proofline-lead as the main-session method where proofline is installed. It takes a request through intake, classification, branch and baseline, discovery, design, planning, freeze, one implementer, independent gates, a bounded fix loop and a completion report. The next command is the source of the next step; the lead does not invent a transition. It dispatches specialists on their configured runtimes and consumes their recorded results.

The lead also carries explicit stop points. It relays specialist questions and waits for decisions on supervised plans, destructive operations, scope changes and delivery actions. It never writes product code, records a specialist verdict or approves its own waiver. When the run reaches a stop or completion, it reports the status and the next action from evidence already recorded.

The lead follows the next command at each transition and keeps one implementer active in a run. Specialists return bounded evidence through their submissions. An unrequested defect becomes a finding for a later bounded fix, not a lead-authored code change. At completion the report lists roles, runtimes, gates, waivers and fix rounds from records. A code-done state does not itself authorize commit or delivery. The lead keeps its context light and reads only result headers, findings tables and status output.

Source Skill purpose

How it works

The skill definition states no numbered steps.

Source Skill definition

Key rules

  • Run the next command before each lifecycle step and do exactly what it says; do not skip, reorder or invent a transition. When next and the skill disagree, next wins and the disagreement is reported.

    Source Lead hard rules

  • Leave product code, tests and fixes to the implementer, including when a run is not active.

    Source Lead hard rules

  • Accept verdicts only through specialist records; the lead cannot record gates or edit their results. A rejected record is not repaired; the work is dispatched again.

    Source Lead hard rules

  • Never ask the user to run agent.sh or another internal command; the Lead runs commands and the user gives a plain reply.

    Source Lead hard rules

  • Require explicit user approval in the current message for commit, push and pull request after all gates for the current content pass.

    Source Lead hard rules

Output and gate

Output

  • At a stop point, return a lead status block with findings and the user's next action.

    Source Output

  • At completion, write a report from recorded roles, runtimes, gates, fixes and findings.

    Source Output

Gate

  • The lead records no gate; code completion requires every required gate and current evidence to pass.

    Source Gate rule

Source

Source Skill definition

← All skills