Skip to content

aiskillstore/marketplace / skills/1102tools-dev/ot-project-description-builder

ot-project-description-builder

Verified

Trigger for: OT project description, OTA scope, Research OT under 10 U.S.C. 4021, Prototype OT under 10 U.S.C. 4022, follow-on production scope under 10 U.S.C. 4022(f), milestone-based project scope, prototype objective, phase or go/no-go structure, BAA white-paper conversion, SOO conversion, transition planning, or scope reduction before OT cost analysis. Produce a milestone-based .docx plus a separate chat-only handoff to OT Cost Analysis. Use TRL only when the project supports it. Never place cost estimates, funding profiles, contribution calculations, prices, or skill-routing content in the document. Do NOT use for a FAR SOW/PWS, OT cost analysis, IGCE, agreement terms and conditions, or an Agreements Officer's authority, participant-status, successful-completion, or follow-on determination.

testing.md

View on GitHub
# OT Project Description Builder: Testing Record

> Historical record from April 2026. References below to a Path D "competition commitment," Prototype OT authority under 10 U.S.C. 4021, or automatic contribution treatment are preserved as test history but are legally superseded. See `test.md` for the August 2026 modernization results and current gates.

## The bottom line

One testing wave in April 2026 across six end-to-end scenarios on Claude Opus 4.7 shipped 15 patches to the OT Project Description Builder. All 15 patches validated on regression across three post-patch runs that exercised untested territory (software prototype, consortium, path D competition commitment, research OT, traditional prime, and Workflow C scope reduction). No regressions, no new universal gaps. All three workflows (A full build, B document conversion, C scope reduction) are now covered.

The skill reliably produces 10 USC 4021 and 4022 compliant project descriptions across prototype, research, and production-follow-on paths, with milestone-based payment structures, TRL-mapped phase progression, data rights handling, cost-sharing path selection, and a separate chat-only milestone handoff table for the OT Cost Analysis.

## Scenarios tested

| # | Scenario | Authority | Performer path | Workflow |
|---|---|---|---|---|
| 1 | Ultra-lazy drone prototype | 10 USC 4021 | NDC | A (full build) |
| 2 | Tethered surveillance drone, richly specified | 10 USC 4021 | NDC, 4022(d)(1)(A) | A |
| 3 | FAR-flavored SOW conversion with CLIN/cost contamination | 10 USC 4021 | Traditional, 4022(d)(1)(C) | B (conversion) |
| 4 | NSTXL-brokered BAA white paper, software prototype | 10 USC 4021 + 4022(f) | Traditional with competition commitment, 4022(d)(1)(D) | B |
| 5 | Cold-weather battery chemistry research | 10 USC 4021 (research) | Traditional, no cost-share required | A |
| 6 | Autonomous resupply prototype, $18M priced, $11M approved, descope to fit budget | 10 USC 4021 | Traditional, 4022(d)(1)(C) | C (scope reduction) |

All six runs produced contract-file-ready .docx outputs, emitted the Milestone Handoff Table as chat output only, and stopped at the Phase 2 Invocation Gate for user "proceed" before document generation.

## What the skill got right on every run

- Four-question Acquisition Context Intake up front (OT type, performer type, TRL entry/exit, consortium/direct)
- Phase 2 Invocation Gate held. No self-approval observed on any run after the gate patch landed.
- Milestone Handoff Table presented as chat-only markdown, never embedded in the .docx or saved as a separate file
- Section 12 Cost-Sharing conditionally omitted for NDC, SB, competition-commitment paths, and for 10 USC 4021 research authority (where 4022(d) is statutorily inapplicable)
- No FAR clause references leaked into the body. The one permitted exception (DFARS 252.227-7013/7014 cited as data rights taxonomy framework under OT tailoring) handled correctly.
- No cost figures, CLINs, FTE counts, or SOC codes in any document body
- Traditional SOW-to-OT conversion correctly stripped CLIN structure, FAR 52 clauses, QASP/AQL language, and dollar amounts while preserving scope intent

## Patches shipped in this wave

Patches were applied in two sub-groups. Group 1 was cross-shipped from the SOW/PWS Builder Wave 2 patch set. Group 2 was derived from three live OT runs and validated against two subsequent runs covering untested territory.

### Group 1: Cross-shipped from SOW/PWS Wave 2 (5 patches)

| Patch | Trigger |
|---|---|
| Phase 2 Invocation Gate with "DO NOT self-approve" language | Universal pattern in document-generating skills |
| Anti-redundancy rule (do not re-ask questions the user's prompt already answers) | Universal pattern |
| AskUserQuestion guidance for Phase 1 intake batching | Matches claude.ai web chat behavior |
| UNCONDITIONAL RULE requiring Milestone Handoff Table emission | Mirrors SOW/PWS unconditional handoff pattern |
| Section ordering prescriptive with explicit renumbering rule for omitted conditional sections | Prevents drift across runs |

### Group 2: OT-specific universal gaps surfaced in Tests 1-3 (10 patches)

| Patch | Section affected | Trigger |
|---|---|---|
| Corrected NDC definition to CAS full-coverage test under 10 USC 3014, removed the incorrect "$500K+ in prior year" threshold | Intake Q2 | Test 2 flagged the factual error; verified against statute |
| Conditional docx generator path (use `/mnt/skills/public/docx/SKILL.md` on claude.ai sandbox, fall back to python-docx on Claude Code and local installs) | Phase 2 opening | Test 2 flagged the sandbox-only path; all three tests had workarounds |
| Split canned handoff closing message into two variants (cost-share and no-cost-share) | Phase 3 handoff | The single version incorrectly promised "apply cost sharing" on NDC, SB, and competition paths |
| Required Document Review Checklist emission in chat (with per-item pass or attention status) before the Milestone Handoff Table | Phase 3 | All three initial runs ran the checklist mentally and moved on |
| Go/no-go gate carry-through rule (phase-boundary milestones must carry Block 2 Q7 criteria into both Section 4 and Section 5) | Phase 1 milestone derivation | Test 1 flagged; criteria collected but not surfaced in the document |
| Defined "Est. Duration" column semantics (calendar months from prior milestone) | Phase 3 handoff table | Ambiguous durations produce 3x+ cost-analysis labor loading deltas |
| Cost data stripping rule for Workflow B (strip source dollar figures from body, pass to handoff as informational only) | Phase 0 Document Intake | Test 3 needed this; skill was silent |
| Cost-share arithmetic disambiguation for Workflow B + 10 USC 4022(d)(1)(C) | Phase 0 Document Intake | Whether a stated total is government share or total agreement value is ambiguous in most SOWs and produces materially different government obligations |
| Performance placeholder closeout mechanism for [TBD] thresholds in Section 3 Objectives, with default bilateral-modification disposition at Phase 1 CDR | Phase 2 Section 14 | Multiple runs invented disposition mechanisms; standardized here |
| Body cross-references must use section titles rather than section numbers | Phase 2 Section Structure | Prevents broken refs when conditional sections are omitted and subsequent sections renumber |

### Regression validation (Tests 4, 5, and 6)

Tests 4, 5, and 6 exercised untested territory after all 15 patches landed. They were designed specifically to stress parts of the skill that Tests 1-3 didn't touch.

- **Test 4 (BAA conversion, consortium, path D, software, late TRL 5-7):** Every Group 2 patch that could fire did fire correctly. The conditional handoff message used the no-cost-share variant. The Document Review Checklist emitted as 11 items in chat. Cross-reference patch held (no numbered refs in body). Source-doc cost stripping wrote "None. White paper contained no cost figures" to the handoff, demonstrating the rule holds even when there is nothing to strip.
- **Test 5 (Research OT, traditional prime, 30 months):** Skill correctly identified that 10 USC 4022(d) cost-sharing paths do not apply to 10 USC 4021 research authority and omitted Section 12. Cost-share arithmetic question did not fire because the statutory framework makes it inapplicable (the correct behavior, not a miss). Go/no-go criteria carried through as placeholder thresholds with stated disposition.
- **Test 6 (Workflow C scope reduction, traditional prime, path C cost-share):** Skill correctly routed to Workflow C. The cost-share arithmetic disambiguation patch, originally written for Workflow B, generalized cleanly to Workflow C and fired as a blocking question with three specific arithmetic interpretations (total value, government share, or pre-share performer quote) and correct math for each. The source-doc cost stripping rule also generalized: original $18M baseline and $11M approved funding carried into the Milestone Handoff Table as "informational only, do not anchor should-cost estimate" rather than leaking into the regenerated document body. Document Review Checklist, Phase 2 Invocation Gate, conditional handoff message (cost-share variant), and [TBD] placeholder closeout all fired correctly. Model also showed useful domain judgment beyond skill rules (flagged NTC as TRL 7 operational and redirected to YPG for TRL 6 relevant environment, presented descope trade-offs as ranked packages A/B/C with risk levels). Those judgment moves were not patch-worthy.

No regressions observed across any of the three regression tests. No new universal gaps surfaced.

**Cross-workflow patch generalization.** Two Group 2 patches originally written for Workflow B (cost-figure stripping, cost-share arithmetic disambiguation) generalized to Workflow C without modification. This is the pattern we want: universal rules that apply wherever source cost data and path (C) math appear, not workflow-specific hedges.

## What was not tested

- **Production follow-on OT as primary workflow.** Tests 2 and 4 included 10 USC 4022(f) provisions as an option, but neither run structured the agreement as a production follow-on.
- **Consortium-specific templates.** Test 4 used NSTXL as the broker; DIU, AFWERX, NavalX, SOSSEC, and MTEC consortium variations have not been exercised.
- **Small business performer path (10 USC 4022(d)(1)(B)).** Tests covered NDC (path A), traditional with competition commitment (path D), traditional with cost-share (path C), and research (no 4022(d) applicability). Small-business significant participation was not tested.
- **Multi-performer consortia.** All tests used single performers or single primes with no subs.
- **Deep classified prototype contexts.** No SAP, SAR, or ICD 705 variants were exercised.
- **Hardware prototype beyond drone and ground vehicle.** Tests 1, 2, and 3 all touched airborne or ground-mobility prototypes. Directed energy, hypersonics, biotech, cyber tooling, and AI/ML prototypes beyond the Test 4 software case have not been exercised.

Users in these contexts should expect to validate outputs more carefully and may encounter edge cases this wave did not surface.

## Testing methodology

Six live runs on Claude Opus 4.7 via claude.ai web chat and Claude Code, the same environments the skill's end users run in.

Evaluator grading was done by a separate Claude Opus 4.7 instance with access only to the skill text and the worker's final output transcript. Grading focused on three questions per run:

1. Did each patch under test fire correctly?
2. Did the output surface any new structural gap not covered by existing patches?
3. If a gap was observed, is it a universal pattern across the skill's full usage profile, or a one-off of model judgment on a specific input?

Only gaps classified as universal structural patterns produced new patches. Drone-specific feedback (unit count defaults, airworthiness prompting), DoD-specific feedback (CMMC, ITAR), and over-engineering suggestions (sentinel markers, trigger-phrase hardening) were deliberately skipped to prevent bloat.

Skill line count: 408 before the wave, 440 after all 15 patches (+32 lines net on a skill that runs hundreds to thousands of times).

## Patches: shipped in this wave

Skill version lines: 408 before patches, 440 after. Ceiling remains 1,000.

All 15 patches are live in the current SKILL.md.

---

**Testing Methodology**

Evaluator: James Jenrette (1102tools) and Claude Code Opus 4.7 (1M context window, max effort mode, Claude Max 20x subscription).

Worker model tested: Claude Opus 4.7 on claude.ai web chat and Claude Code, the same environments the skill's end users run in.

Wave: 6 runs, 15 patches shipped, 3 post-patch regression runs with zero new gaps surfaced. All three workflows (A, B, C) covered.

Date: April 2026.

Skill: ot-project-description-builder. Source: github.com/1102tools-dev/federal-contracting-skills. License: MIT.
canvas-designanthropics/skillsCreate beautiful visual art in .png and .pdf documents using design philosophy. You should use this skill when the user asks to create a poster, piece of art, design, or other static piece. Create original visual designs, never copying existing artists' work to avoid copyright violations.Stars: 198kUpdated Claude CodeCodexVerifiedInstall
docxanthropics/skillsUse this skill whenever the user wants to create, read, edit, or manipulate Word documents (.docx files) or Word templates (.dotx files). Triggers include: any mention of 'Word doc', 'word document', '.docx', '.dotx', or requests to produce professional documents with formatting like tables of contents, headings, page numbers, or letterheads. Also use when extracting or reorganizing content from .docx or .dotx files, inserting or replacing images in documents, performing find-and-replace in Word files, working with tracked changes or comments, or converting content into a polished Word document. If the user asks for a 'report', 'memo', 'letter', 'template', or similar deliverable as a Word or .docx file, use this skill. Do NOT use for PDFs, spreadsheets, Google Docs, or general coding tasks unrelated to document generation.Stars: 198kUpdated Claude CodeCodexVerifiedInstall
pdfanthropics/skillsUse this skill whenever the user wants to do anything with PDF files. This includes reading or extracting text/tables from PDFs, combining or merging multiple PDFs into one, splitting PDFs apart, rotating pages, adding watermarks, creating new PDFs, filling PDF forms, encrypting/decrypting PDFs, extracting images, and OCR on scanned PDFs to make them searchable. If the user mentions a .pdf file or asks to produce one, use this skill.Stars: 198kUpdated Claude CodeCodexVerifiedInstall
pptxanthropics/skillsUse this skill any time a .pptx or .potx file is involved in any way — as input, output, or both. This includes: creating slide decks, pitch decks, or presentations; reading, parsing, or extracting text from any .pptx or .potx file (even if the extracted content will be used elsewhere, like in an email or summary); editing, modifying, or updating existing presentations; combining or splitting slide files; working with templates (.potx), layouts, speaker notes, or comments. Trigger whenever the user mentions "deck," "slides," "presentation," or references a .pptx or .potx filename, regardless of what they plan to do with the content afterward. If a .pptx or .potx file needs to be opened, created, or touched, use this skill.Stars: 198kUpdated Claude CodeCodexVerifiedInstall