agentleFS
chakmarebelchakmarebel/federal-proposal-copilot/.claude/skills/capture-submission/SKILL.md

capture-submission

Skill4 starsChanged 3 months agoMITView on GitHub

Use this skill to snapshot a proposal's auto-generated draft and final submitted version into the calibration corpus, with edit notes. Run twice per proposal — once before manual editing (snapshot the AI's output) and once after submission (capture the final + notes). Builds the data the framework needs to learn from real edits and close the gap between auto-generation and submission-ready.

The file

We split this skill into the rules it gives an agent

11 rules. About 0 are common, 3 are distinctive, and 8 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈3

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 8

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Run this skill twice per proposal:

    line 17
  2. Distinctive:

    Read in this order:

    line 26
  3. Very specific:

    When corpus/calibration/<slug>/ does not exist:

    line 47
  4. Very specific:

    Read working/proposal-type.md — confirm a proposal is active.

    line 49
  5. Very specific:

    If working/compliance-matrix.md exists, copy it to corpus/calibration/<slug>/auto-draft/_compliance-matrix.md for context.

    line 52
  6. Very specific:

    If working/graphics-brief.md exists, copy it to corpus/calibration/<slug>/auto-draft/_graphics-brief.md.

    line 53
  7. Very specific:

    Create corpus/calibration/<slug>/manifest.json from _template/manifest.json.example with these fields filled in:

    line 54
  8. Very specific:

    When corpus/calibration/<slug>/auto-draft/ exists but final-submitted/ does not:

    line 86
  9. Very specific:

    When auto-draft/, final-submitted/, and edit-notes.md (filled out) all exist:

    line 114
  10. Distinctive:

    If still has placeholders → print reminder, do NOT mark complete.

    line 117
  11. Very specific:

    Never overwrite an existing auto-draft: If auto-draft/ already exists, you're past Phase 1.

    line 154

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Capture Submission Skill

Purpose

Snapshot the (auto-generated draft, final submitted version, edit notes) tuple for a submitted proposal into corpus/calibration/. This is the data layer that lets the framework learn from your edits and improve over time.

When to use

Run this skill twice per proposal:

  1. After /proposal-writer completes, before you start editing for submission. This snapshots the AI's actual output as a baseline. If you skip this step and start editing first, the auto-draft is gone — you can't reconstruct it.
  2. After you submit the proposal. This captures the final version + opens the edit-notes template for you to fill out.

The skill detects which phase you're in based on what already exists in corpus/calibration/<slug>/.

Inputs

Read in this order:

  1. working/proposal-type.md — for proposal_type and to confirm a proposal is active
  2. corpus/calibration/<slug>/manifest.json (if exists) — to determine current phase
  3. drafts/ — source for auto-draft snapshot (Phase 1)
  4. final/ — source for final-submitted snapshot (Phase 2). Look in final/docx/ first (most reliable post-export source), fall back to drafts/ if user submitted markdown directly
  5. corpus/calibration/_template/ — template files for new entries

Phase detection

The skill chooses behavior based on existing state:

StateDetectionPhaseAction
corpus/calibration/<slug>/ does not existFirst runPhase 1: pre-edit snapshotSnapshot drafts/ → auto-draft/, init manifest.json, copy edit-notes.md.template
corpus/calibration/<slug>/auto-draft/ exists, final-submitted/ does notSecond runPhase 2: post-submit captureSnapshot final/ (or drafts/ if user names the source) → final-submitted/, update manifest.json, prompt user to fill edit-notes.md
Both auto-draft/ and final-submitted/ exist, edit-notes.md is filled out (more than the template defaults)Third runPhase 3: completeUpdate manifest.json to status=complete, suggest running /red-team-review --mode=lessons-learned
Both exist, edit-notes.md is still template defaultsReminderReminderPrint "edit-notes.md still has template placeholders — please fill it out for the entry to be useful"

Phase 1: Pre-edit snapshot

When corpus/calibration/<slug>/ does not exist:

  1. Read working/proposal-type.md — confirm a proposal is active. Extract type_id and customer (from working/customer-profile.md if available).
  2. Create directory: corpus/calibration/<slug>/
  3. Copy drafts/*.md → corpus/calibration/<slug>/auto-draft/ (preserve filenames). Skip empty drafts and drafts/.gitkeep.
  4. If working/compliance-matrix.md exists, copy it to corpus/calibration/<slug>/auto-draft/_compliance-matrix.md for context.
  5. If working/graphics-brief.md exists, copy it to corpus/calibration/<slug>/auto-draft/_graphics-brief.md.
  6. Create corpus/calibration/<slug>/manifest.json from _template/manifest.json.example with these fields filled in:
    {
      "schema_version": "calibration-manifest.v1",
      "slug": "<slug>",
      "proposal_type": "<type_id>",
      "customer": "<customer name or null>",
      "captured_pre": "<ISO 8601 timestamp now>",
      "captured_post": null,
      "submitted_date": null,
      "status": "pre-edit",
      "outcome": null,
      "edit_hours_estimate": null,
      "notes": ""
    }
  7. Copy _template/edit-notes.md.template → corpus/calibration/<slug>/edit-notes.md and substitute [Proposal Slug] with the actual slug.
  8. Append to working/activity.md:
    ## YYYY-MM-DD HH:MM — capture-submission [phase=pre] — auto-draft snapshot → corpus/calibration/<slug>/auto-draft/
  9. Print to user:
    Auto-draft snapshot captured to corpus/calibration/<slug>/auto-draft/.
    
    Continue editing in drafts/ (or final/ after running /export-proposal).
    When you submit, re-run /capture-submission to capture the final version
    and fill out edit-notes.md.

Phase 2: Post-submit capture

When corpus/calibration/<slug>/auto-draft/ exists but final-submitted/ does not:

  1. Locate the final submitted source. Priority order: a. final/docx/ — if exists and contains files, this is the most reliable post-export source. Note: these are .docx files. Use the anthropic-skills:docx skill to extract markdown text from each, OR if the user has run /export-proposal and kept the markdown source intact, use the source markdown. Prefer markdown for diffability. b. final/<format>/ for any format that has files c. drafts/ — fallback if user submitted markdown directly without exporting d. If none of the above, prompt the user: "I can't find the final submitted version. Where is it? (path or 'paste here')"
  2. Copy the located files → corpus/calibration/<slug>/final-submitted/ (preserve section filenames; if from .docx, write extracted markdown).
  3. Update manifest.json:
    • captured_post: ISO 8601 timestamp now
    • submitted_date: prompt user "What date did you submit? (YYYY-MM-DD, or press Enter for today)"
    • status: "post-submit"
  4. Prompt user: "Open corpus/calibration/<slug>/edit-notes.md and fill it out. Five minutes. The most valuable section is #3 (top 3 systemic patterns)."
  5. Append to working/activity.md:
    ## YYYY-MM-DD HH:MM — capture-submission [phase=post] — final captured + edit-notes ready → corpus/calibration/<slug>/
  6. Print to user:
    Final version captured to corpus/calibration/<slug>/final-submitted/.
    
    Next step: open corpus/calibration/<slug>/edit-notes.md and fill it out
    (5 min). Then re-run /capture-submission one more time to mark the entry
    complete and unlock /red-team-review --mode=lessons-learned for this entry.

Phase 3: Mark complete + suggest review

When auto-draft/, final-submitted/, and edit-notes.md (filled out) all exist:

  1. Detect "filled out" by checking that edit-notes.md has more than just the template placeholders. Heuristic: at least one of sections 1, 3, or 4 has non-placeholder content.
  2. If still has placeholders → print reminder, do NOT mark complete.
  3. If filled out:
    • Update manifest.json:
      • status: "complete"
      • edit_hours_estimate: extract from edit-notes.md section 1 if user filled it in
      • outcome: prompt "Outcome (awarded/not-awarded/pending/withdrawn, or press Enter for pending)"
    • Append to working/activity.md:
      ## YYYY-MM-DD HH:MM — capture-submission [phase=complete] — entry finalized → corpus/calibration/<slug>/
    • Print to user:
      Calibration entry complete: corpus/calibration/<slug>/
        auto-draft/         (N files)
        final-submitted/    (N files)
        edit-notes.md       (filled)
        manifest.json       (status=complete)
      
      Recommended next step:
        /red-team-review --mode=lessons-learned
      
      This will diff auto-draft vs final-submitted, ingest your edit notes,
      and produce framework improvement proposals in
      reviews/lessons-learned-<slug>.md.

Output paths

  • corpus/calibration/<slug>/auto-draft/<section>.md (Phase 1)
  • corpus/calibration/<slug>/auto-draft/_compliance-matrix.md (Phase 1, if exists)
  • corpus/calibration/<slug>/auto-draft/_graphics-brief.md (Phase 1, if exists)
  • corpus/calibration/<slug>/manifest.json (Phase 1, updated each phase)
  • corpus/calibration/<slug>/edit-notes.md (Phase 1, populated by user, finalized in Phase 3)
  • corpus/calibration/<slug>/final-submitted/<section>.md (Phase 2)

Critical Rules

  • Never overwrite an existing auto-draft. If auto-draft/ already exists, you're past Phase 1. Detect the phase and act accordingly. Re-snapshotting the auto-draft after edits would destroy the calibration signal.
  • Never overwrite an existing final-submitted. Same logic.
  • Never edit edit-notes.md content yourself — that's the user's reflection. The skill only creates the file from template; the user fills it.
  • No PII in manifest.json. Customer name and outcome are OK; specific evaluator names, contact info, or pricing data are not. The manifest is gitignored along with the per-proposal directory, but defense-in-depth.
  • Read-only on the corpus once a phase is complete. Each phase appends or fills in fields; no phase rewrites prior phases.

Cross-reference

  • Discipline + structure: corpus/calibration/README.md
  • Templates: corpus/calibration/_template/
  • Downstream consumer: red-team-review --mode=lessons-learned reads completed corpus entries and produces framework improvement proposals
  • Future cross-proposal analysis: a /recalibrate-framework skill (not yet built) will read multiple corpus entries to surface patterns visible only across proposals

Activity Trail

On completion, append to working/activity.md in the proposal workspace (NOT the corpus directory):

## YYYY-MM-DD HH:MM — capture-submission [phase=<pre|post|complete>] — <one-line summary> → corpus/calibration/<slug>/

This makes the calibration capture visible in /status output and the dashboard.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.