docs: close SYSTEM_PROMPT process gaps to prevent agent halting

[avoidwork]

Description

Audit and fix process gaps in prompts/SYSTEM_PROMPT.md that could cause the agent to stop processing pending work and require a 'continue' prompt from the user.

Type of Change

• Documentation update

Changes

1. Universal no-pausing rule (new directive 9) — explicit instruction not to pause between tool calls or at natural breakpoints during any execution path, not just todo queues.
2. Defined 'genuinely blocked' — concrete definition in ambiguity handling: only zero viable paths forward qualifies. Minor ambiguities are not blockers.
3. Rewrote examples — removed collaborative handoff language ("send it over and we will shape it together") that modeled pausing behavior.
4. Failure escalation — when all recovery paths are exhausted, report and continue. Never stop the entire workflow because one step failed.
5. Implied sub-tasks (new directive 10) — execute the full chain the user expects (code → test → commit → verify), not just the primary deliverable.
6. Reconciled contradiction — clarified that "Answer what was asked" does not override "Complete Implied Sub-Tasks."
7. Removed "decide whether to continue" — TODO and AGENT sections now say "continue" unconditionally instead of giving the agent discretion to stop.
8. Scoped "Show your work" — applies to final answers, not intermediate tool operations.
9. Verbosity guard — terse during technical execution; persona in delivery, not execution log.
10. Impossible/wrong requests — proceed with a warning and safe interpretation instead of stalling.

Testing

N/A — documentation-only changes.

Coverage

• 100% line coverage maintained

Checklist

• npm run lint passes
• Tests pass with 100% line coverage
• No forbidden patterns used
• Conventional Commit style applied

Rule 5.4 Compliance

PR template section filled out in full — no blank sections.