Skip to content

agent-completion-discipline

Agent Completion Discipline — Commit Before Summary

Section titled “Agent Completion Discipline — Commit Before Summary”

Resolves: unity#74 (tail-of-thought stop — agent exits without committing pending work at >150K tokens)

COMMIT BEFORE YOU SUMMARIZE. Any agent that has made edits, written files, or completed implementation work MUST commit + push those changes BEFORE composing any narrative summary, wrap-up, or report.

Order of operations (mandatory):

  1. Dispatch all pending Write operations
  2. git add + git commit + git push
  3. THEN compose the summary / report / wrap-up message

150K token checkpoint (MANDATORY for sub-agents): At ~150K context tokens, STOP all investigation and immediately:

  1. Check git status — if any staged or unstaged changes exist, commit them NOW
  2. Dispatch any pending Write operations before reading more files
  3. Only resume investigation or compose a summary AFTER the commit lands

“Let me check one more thing before committing…” at >150K tokens.

That sentence is the symptom. Interrupt it. Commit first. Investigate after.

7/8 sub-agent stops in the 2026-05-14 BackpackCrawler session (unity#74) occurred because agents reached 168–212K tokens mid-investigation and exited as completed without committing already-made edits. Recovery required separate commit-only finisher agents, doubling spawn cost per phase.

Validation (would this rule have prevented unity#74?)

Section titled “Validation (would this rule have prevented unity#74?)”

Yes — all 7 failures shared the same pattern: investigation continued past 150K tokens with uncommitted edits. The 150K checkpoint with mandatory git status check would have forced commits at the right moment in all 7 cases.