Skip to content

feat: accept the documents users already have, not JSON - #14

Merged
mikeastock merged 3 commits into
mainfrom
feat/proposal-builder-plain-inputs
Oct 14, 2026
Merged

mikeastock merged 3 commits into
mainfrom
feat/proposal-builder-plain-inputs

Conversation

@mikeastock-bot

@mikeastock-bot mikeastock-bot commented Oct 14, 2026 •

Copy link
Copy Markdown

Users hand an agent PDFs, Word files, spreadsheets, and emails, never JSON. Across the three skills that asked for structured input, JSON is now the agent's internal working file, built from the user's material as-is and edited when the user asks for a change in plain words.

proposal-builder (1.1.0)

  • Accepts any format; asks for missing material in contractor terms (project list, resumes, carrier EMR letter, OSHA 300A, approach write-ups). Unreadable files are named and re-requested, never guessed.
  • proposal.json is agent-owned. Users get the PDF and compliance-check.md; changes ("Sam is out, use Jordan") are edited in, re-rendered, and re-checked.
  • Library sources are dated from the document itself, not file mtimes. A new library is written as editable Markdown with a Last verified: line.
  • Sample library: {projects,people,safety}.json → .md. Compliance check output is unchanged.

bid-leveling (1.1.0)

  • Bids and scope sheet come in as PDFs, emails, spreadsheets, or pasted text, instead of "as Markdown, or ask for Markdown". Summary no longer says "Markdown bids in".
  • bid-data.json is explicitly the agent's working file; plugs can come from a message or a budget sheet.

rfp-intake (1.1.0)

  • No longer asks for company-profile.json. The agent asks for the gate facts in plain terms or reads them from the bonding letter, COI, license list, or scorecard, then writes and maintains company-profile.json itself.
  • Script validation errors are fixed by the agent, not relayed to the user.

Marketplace 0.13.0, catalog.json rebuilt. workforce-planning already worked from the files ops teams keep and is unchanged.

Note

Accept native documents instead of JSON across bid-leveling, proposal-builder, and rfp-intake skills

  • Updates all three skills (v1.1.0) to take PDFs, Word files, spreadsheets, emails, and pasted text as input instead of asking users to convert to JSON or a template
  • JSON working files (proposal.json, intake-data.json, company-profile.json, bid-data.json) are now assistant-owned; the assistant builds and edits them, fixes validator errors internally, and only delivers the proposal PDF, compliance check, or plain-language results
  • Company-library samples and fallback output migrate from JSON to Markdown files (projects.md, people.md, safety.md) with a Last verified line, in library-and-prior-proposals.md
  • Unreadable scans, photos, or password-protected files trigger a request for a readable copy; the skills forbid guessing numbers or exclusions from unreadable input
  • Behavioral Change: users who previously supplied JSON libraries must now supply source documents or plain Markdown; deleted samples projects.json, people.json, safety.json in skills/proposal-builder/samples/company-library/ are replaced by their .md equivalents

Macroscope summarized bea6495.

@mikeastock-bot mikeastock-bot changed the title feat(proposal-builder): accept the documents users already have, not JSON feat: accept the documents users already have, not JSON Oct 14, 2026
@mikeastock
mikeastock merged commit f01e9af into main Oct 14, 2026
1 check passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants