<objective> Run multiple SHINE skills back-to-back, piping the output of each into the next via a shared scratchpad. Lets the user build ad-hoc workflows without hard-coding a new skill for every combination.

Example:

/shine-pipeline shine-lead-enrich shine-cold-email shine-followup-draft

→ enrich leads → draft cold emails → draft follow-ups, all in one command, all sharing context. </objective>

<execution_context> @~/.claude/shine/references/ui-brand.md </execution_context>

<context> Arguments: an ordered list of skill names (2 or more). Each must exist at `~/.claude/skills/<name>/SKILL.md`.

Shared scratchpad: .planning/pipeline-<timestamp>.md in the current working directory. Each step appends its summary + any structured output to this file, and the next step reads it before starting.

No magic data transform — skills that naturally produce markdown/tables compose best. Skills that only mutate the filesystem still work (they just contribute a brief summary to the scratchpad). </context>

<execution_flow>

Validate arguments. Require at least 2 skill names. Check each has a SKILL.md. If any is missing, list them and stop.

Initialize scratchpad.

TS=$(date +%Y%m%d-%H%M%S)
SCRATCH=".planning/pipeline-$TS.md"
mkdir -p .planning
printf "# SHINE Pipeline — %s\n\n**Steps:** %s\n\n" "$TS" "$*" > "$SCRATCH"

Run each skill in order. For step i (skill S_i):
- Append ## Step i: S_i header to scratchpad.
- Invoke the skill via Skill(S_i) with the scratchpad path passed as context.
- After it returns, append its result summary (last structured output block) to the scratchpad under ### Output.
- If the skill fails or returns empty, pause and ask the user: "Step i (S_i) returned no output. Continue with next step, retry, or abort?"
Final summary. After the last step, print:
- Pipeline duration.
- List of steps with ✅ / ⚠️ per step.
- Path to the scratchpad.
Cleanup. Leave the scratchpad in place — it's a reviewable artifact, not a temp file. </execution_flow>

<guardrails> - **Max 8 steps** per pipeline — more than that and you're building a skill, not a pipeline. Hard-stop with a clear message. - **No silent skill substitution.** If a skill name is misspelled, stop; don't fuzzy-match. - **Never chain destructive skills without confirmation.** If any step's `allowed-tools` includes `Bash` or a skill name contains `send`, `publish`, `deploy`, `commit`, or `push`, ask for explicit approval before running that step. - **Context budget awareness.** Before each step, estimate scratchpad size; if >50 KB, warn the user and offer to summarize before continuing. - **Never auto-apply memory changes** mid-pipeline. Even if a step proposes memory updates, defer to `/shine-retro` or manual edits. </guardrails>

<error_handling>

Missing skill: list all missing skills at once, exit without starting. Don't run partially.
Step failure: pause, show last 20 lines of output, offer retry/skip/abort.
Scratchpad unwritable: fall back to $TMPDIR/shine-pipeline-<ts>.md and warn.
User abort mid-pipeline: leave scratchpad with **ABORTED at step i** marker so the run is debuggable. </error_handling>

<output_format> Final summary block:

## Pipeline complete — <TS>

| # | Skill | Status | Notes |
|---|-------|--------|-------|
| 1 | shine-lead-enrich | ✅ | 42 leads enriched |
| 2 | shine-cold-email | ✅ | 42 drafts ready |
| 3 | shine-followup-draft | ⚠️ | paused — needs review |

**Scratchpad:** .planning/pipeline-<TS>.md
**Duration:** 3m 12s

</output_format>

<structured_returns> Exit 0 if all steps completed or were user-skipped. Exit 1 if any step was aborted or failed without user resolution. </structured_returns>