issue-lifecycle skill: improve resumption and close-out guidance

Problem Statement

The issue-lifecycle skill's "Resuming a Session" section (SKILL.md lines 109-118) has three gaps that cause agents to corrupt lifecycle state when picking up an issue mid-flight:

1. Wrong resume command. The section says to run swamp model get issue-<N> --json, which returns the model definition (methods, schemas) — not the current phase. The correct command is swamp data get issue-<N> state-main --json.

2. No warning that start resets state. The start method always resets the phase to triaging, even if the issue was at pr_open. An agent that doesn't know this calls start thinking it's idempotent, then gets stuck because the model enforces strict phase transitions.

3. No close-out workflow. When a PR has already merged and the lifecycle just needs to be marked shipped, there's no guidance. The agent ends up manually walking through every intermediate phase transition.

Proposed Solution

Three changes to the skill files:

1. Rewrite "Resuming a Session" in SKILL.md — use the correct swamp data get command, add a warning about start, and include a phase-to-action reference table mapping each phase to what to do next when resuming.

2. Add "Closing Out a Shipped Issue" section in SKILL.md — a short workflow for the common case: check current phase, link PR if needed, call complete.

3. Add start safety warning in references/triage.md — a blockquote after the start method usage in section 2 clarifying that start should only be called once per issue.

Alternatives Considered

A detailed implementation plan with exact line numbers and replacement text is available in the plan file at .claude/plans/twinkly-humming-mango.md in the swamp repo working directory.

Environment

• swamp version: 20260409.030033.0-sha.07dae003
• OS: macOS Darwin 25.3.0