run-cycle

/sws:run-cycle

End-to-end SWS pipeline driver. Re-run safely — completed steps are skipped.

Usage

/sws:run-cycle                                  # full pipeline
/sws:run-cycle --dry-run                        # plan only, no dispatch
/sws:run-cycle --only=cover-letter,disclosure   # subset

Step plan (D9)

Step	Name	Skill dispatched	Skip condition
1	outline	/sws:outline-paper	outline.md exists
2	draft	/sws:draft-paper	_drafts/*.md non-empty
3	revise	/sws:revise-paper	_drafts/*.revised.md newer than canonical drafts
4	review	/sws:review-paper	all 3 _review//report.md present
5	cover-letter	/sws:write-cover-letter	_submission/cover-letter.md exists OR not required
6	disclosure	/sws:disclose-ai-usage	_submission/ai-disclosure.md exists OR not required
7	response	/sws:respond-to-reviewers	no pending _review/round-/reviewer-comments.md

Steps

1. Verify ${PAPER_ROOT}/.sws-project.local.md exists.
2. Parse --dry-run and --only=<csv> flags.
3. Compute the step plan:

   ${CLAUDE_PLUGIN_ROOT}/scripts/sws_python.sh "$PAPER_ROOT" \
       ${CLAUDE_PLUGIN_ROOT}/scripts/sws_run_cycle.py \
       --paper-root "$PAPER_ROOT" --dry-run --json [--only=<csv>]
   

   Capture the JSON plan; print a human-readable summary to the user.
4. If --dry-run was passed, exit 0 here.
5. Otherwise, iterate the planned steps in order. For each step with should_run: true:
   • Determine dispatch kind by step name:

     case "$step_name" in
       outline|draft|revise|review|cover-letter|response)
           # Agent-backed step — emit MANUAL directive for the user-context
           # to dispatch the matching skill. The orchestrating context
           # (typically Claude Code in the user's session) reads the
           # directive and invokes the skill.
           echo "MANUAL: dispatch ${skill_name} (${step_name})"
           ;;
       disclosure)
           # Deterministic script — dispatch directly.
           ${CLAUDE_PLUGIN_ROOT}/scripts/sws_python.sh "$PAPER_ROOT" \
               ${CLAUDE_PLUGIN_ROOT}/scripts/sws_disclosure_writer.py \
               --paper-root "$PAPER_ROOT"
           ;;
     esac
     

6. After all dispatchable steps complete, append a single passport history entry with:
   • phase: submit
   • venue: <target_journal> (from marker)
   • round: <N> (only when the response step was dispatched)
   • change_summary: comma-separated list of step names dispatched
   • next_step: "ready for upload" if all required artifacts exist, else a short TODO list (e.g. "fill {EDITOR_NAME} in cover-letter.md").
7. Print a final summary listing each step's outcome (RUN | SKIPPED | MANUAL).

Stale-detection (R4, D10)

When _submission/cover-letter.md predates a _drafts/*.md source file by more than 5 seconds, the planner flags the step as stale. The orchestrator prints a notice but does NOT auto-regenerate — the user re-runs the specific skill if they want a refresh.

Profile gating

Steps 5 (cover-letter) and 7 (response) check both the RESOLVED_COVER_LETTER_REQUIRED / RESOLVED_DISCLOSURE_REQUIRED resolver fields and the active profile's agent-activation list before dispatch (D8).

Spec source of truth

docs/superpowers/specs/2026-05-30-cycle-12-submission-orchestration-design.md — D2, D9, D10, D14, R4.