Enforcing Complete Output
Markdown--- name: enforcing-complete-output description: Ensures generated content (code, documents, lists) is delivered in full without placeholders, truncation, or shortcuts. Use when a user requests multiple deliverables (files, functions, sections, items) and needs guaranteed completeness, or when output risks being cut off by length limits and must resume cleanly. ---
Before generating anything, count the deliverables explicitly:
"Scope: 5 files requested —
auth.py,models.py,routes.py,tests.py,config.py. Building all 5 in full."
Then generate each one completely. No // ..., no # rest of implementation, no "similarly for the others." If the full output won't fit, stop at a clean boundary and emit:
[PAUSED — 3 of 5 complete. Send "continue" to resume from: routes.py]
Progress tracking checklist (use for any multi-deliverable request):
Scope confirmed: N deliverables identified
- [ ] Deliverable 1: <name> — fully implemented
- [ ] Deliverable 2: <name> — fully implemented
- [ ] ...
- [ ] Cross-check: count matches scope
- Scope — Parse the request and enumerate every distinct deliverable (files, functions, sections, list items). State the count before producing anything.
- Build — Generate each deliverable in full, in order. Treat every item as if it were the only thing being written — no summarizing, no deferring detail.
- Cross-Check — Before finalizing, re-count what was produced against the scope. If any deliverable is missing or incomplete, finish it before ending the turn.
- Token Limit Handling — If length constraints force a stop:
- Stop only at a clean boundary (end of a file, function, or section — never mid-statement, mid-function, or mid-sentence).
- Emit exactly:
[PAUSED — X of Y complete. Send "continue" to resume from: <next deliverable name>] - On resuming, continue directly from that point with zero recap, zero re-summary, zero repeated content.
Example 1:
Input: "Write 4 Python utility functions: slugify, chunk_list, deep_merge, retry_decorator."
Output: All four functions written with full bodies, docstrings, and edge-case handling — no # implementation here placeholders. Ends with confirmation: "4 of 4 complete."
Example 2: Input: "Generate 10 React components for a dashboard." (output too large for one response) Output: Components 1–6 fully coded, then:
[PAUSED — 6 of 10 complete. Send "continue" to resume from: Component 7 - RevenueChart]
On "continue": Component 7 begins immediately, no restatement of components 1–6.
Example 3: Input: "List and fully explain 8 design patterns." Output: All 8 patterns, each with a complete explanation and code example — never "the remaining patterns follow a similar structure."
- State the deliverable count explicitly at the start — this creates a self-check anchor.
- Treat each deliverable as an independent full task, not a continuation to be abbreviated.
- Choose pause points at natural boundaries (end of file/function/section), never mid-construct.
- When resuming, re-state only the next deliverable's name — nothing else.
- If a deliverable is inherently large (e.g., a long config file), still complete it fully before moving to the next, even if that consumes most of the remaining space.
- Don't write "// rest remains unchanged" or "similarly for the others" — this violates the core mandate even if well-intentioned.
- Don't silently skip a deliverable because it seems repetitive — repetitive ≠ omittable.
- Don't pause mid-function or mid-item to save space — always find the nearest clean breakpoint.
- Don't re-summarize prior completed work when resuming — this wastes space and violates "zero recap."
- Don't estimate completion ("most of the functions are done") — always give an exact X of Y count.