1102tools Agent Skills

Agent skills for federal acquisition deliverables: SOW / PWS, IGCEs, OT project descriptions, OT cost analyses. Six orchestration skills that handle scope decisions, cost buildup, FAR citations, and document generation.

Website: 1102tools.com

Download the universal setup guide (PDF), then drop it into Claude, ChatGPT (Codex), Gemini (Antigravity), Copilot, DeepSeek Harness, Grok, Cursor, opencode, or LibreChat and say what you want installed. The AI reads the guide and walks you through every step: free API keys, the exact config for your platform, restart, verify. If 35 pages is more than your chat will accept, paste in just the section for your platform; every option is written to stand alone. Part 9 is troubleshooting built from errors I actually hit while executing every major install path for real.

Before you build: Not every acquisition capability should be an AI tool. Dozens of potential skills were evaluated and several were intentionally excluded because they cross the line from data assembly into professional judgment. See ai-boundaries.md.

For federal API data (SAM.gov, BLS wages, GSA CALC+ rates, GSA Per Diem, USASpending, eCFR, Federal Register, Regulations.gov), use the companion repo:

federal-contracting-mcps: eight MCP servers covering those APIs, installed with one config block per server in any MCP client.

The two repos work together: MCPs handle data, skills handle deliverables.

The MCP servers these skills call were rebuilt on v2 of the MCP Python SDK and published at 1.0.0. Update them, then read the two notes below, because both affect estimates you may already have produced.

Check any IGCE you built between roughly April and August 2026. The BLS OEWS server was defaulting to a data year that BLS had withdrawn when it published the 2025 estimates. Wage lookups returned empty values rather than an error, which is indistinguishable from a privacy-suppressed cell. All three IGCE Builders and OT Cost Analysis price labor off that server. If an estimate from that window shows missing or suppressed wages, or you worked around them by hand, rebuild it.

The skills in this repo were also carrying the old vintage. They hardcoded May 2024 as the BLS data vintage and computed the wage aging factor from it. Pairing an updated MCP with the old skill would age 2025 wages from a 2024 vintage, adding a full extra year of escalation and overstating every labor line by roughly one year of your escalation rate. The vintage is now May 2025, and each pricing skill carries a standing instruction to confirm it with detect_latest_year() before pricing rather than trusting the constant. Re-download the four pricing skills.

Apologies to anyone who lost time to either one. The MCP-side detail is in each server's changelog.md.

.mcpb bundles are discontinued. They could not be signed in a way Claude Desktop recognizes, so every install raised an untrusted-developer prompt, and they re-resolved dependencies on each launch instead of pinning them. Install via the config block in the companion repo instead.

The five API data-source skills (BLS OEWS, GSA CALC+, GSA Per Diem, SAM.gov, USASpending) were removed from this repo in April 2026. They moved to the MCP companion repo.

Reasons:

1. Deterministic tool calls. MCP servers execute tested Python code. The model does not generate API-call code on the fly. Skills drifted across runs; MCPs do not. Same input, same output.
2. Updated independently of the skills. When an upstream API changes, the MCP ships a fix and you get it on the next launch. Nothing to re-upload. A data-source skill would have to be re-downloaded and re-installed by every user.
3. Less context cost. Tool schemas are ~100 tokens each. The old API skills cost 500-1000 lines of context per run.
4. Production-hardened. Each MCP went through 3-6 audit rounds with live testing against the production API. Roughly 350 bugs fixed during hardening across the eight MCPs.
5. Cross-client support. MCP is an open standard, so the same servers run in Claude Desktop, Claude Code, Codex, Antigravity, and Copilot. This was the strongest argument for the split in April, when skills were Claude-only. It has since weakened: SKILL.md is now read by nearly every runtime too. The other four reasons still hold, and are why the split stays.

The orchestration skills in this repo stay as skills. Their value is decision trees, FAR-compliant narrative, and document generation, not API calls.

Model note: Built and tested with Claude, on Opus. These are dense single-file skills, 10,000 to 18,000 tokens each, and that density is what smaller models struggle with: expect missed steps and wrong outputs. That is a model limitation, not a platform one. See the panel at the top for which runtimes read the format.

"Requires" lists the MCP servers each skill calls at runtime. Install them from the companion repo.

1. Install Claude Desktop, or use any runtime from the panel above.
2. Install the MCPs you need from federal-contracting-mcps. Each server's README has a copy-paste config block; add it to your client config and restart.
3. Add the skill. In Claude Desktop: Customize > Skills > + > Create skill > Upload a skill. Elsewhere, drop the unzipped folder into that runtime's skills directory:

4. Ask naturally. The skill reads its instructions and calls the MCP tools.

The pricing skills need three MCP servers and a Python environment with openpyxl to produce their workbooks. A runtime with no code execution can read them but cannot finish the job.

MIT

Built by James Jenrette, lead systems analyst and contracting officer. Independently developed and not endorsed by any federal agency.