proposal-patcher
Skill4 starsChanged 3 months agoMITView on GitHub
Apply audit findings (Gold Team Weaknesses/Deficiencies and evidence-check CLAIM-UNSUPPORTED markers) as surgical fixes to bound drafts. Never rewrites globally. Preserves voice, structure, and all Strengths. Final stage of the Track B white-paper workflow before export.
In this repository30 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- adversarial-reviewSkill
- capture-demand-signalsSkill
- capture-intentSkill
- capture-portal-structureSkill
- capture-scorecardSkill
- capture-submissionSkill
- competitor-assessmentSkill
- compliance-checkSkill
- customer-intelSkill
- dashboardSkill
- evidence-checkSkill
- export-proposalSkill
- import-from-captureSkill
- narrative-spineSkill
- new-proposalSkill
- opportunity-quick-lookSkill
- past-performanceSkill
- pricing-analystSkill
- proposal-editorSkill
- proposal-graphicsSkill
- proposal-managerSkill
- proposal-solution-architectSkill
- proposal-storyboardSkill
- proposal-writerSkill
- red-team-reviewSkill
- setup-companySkill
- statusSkill
- submission-summarySkill
- technical-reviewSkill
The file
Show its rules and text
We split this skill into the rules it gives an agent
23 rules. About 0 are common, 15 are distinctive, and 8 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈15
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
- Very specific: line 21
Run after both /evidence-check and /red-team-review --mode=gold have written their findings.
- Very specific: line 33
Do not run /proposal-editor before this skill.
- Distinctive: line 50
Before touching any draft, build a complete patch list.
- Distinctive: line 58
Then stop and confirm: Ask the user: "This is the patch list — N findings.
- Distinctive: line 62
After confirmation, apply the patches.
- Distinctive: line 73
Write in the section's existing voice — match cadence, sentence length, register.
- Distinctive: line 74
Do not write a new section.
- Very specific: line 75
If a Deficiency cannot be addressed without fabricating a capability or past performance, flag it for the user: <!-- PATCH-BLOCKED: …
- Distinctive: line 81
If the fix would require a significant structural change (reordering sections, adding >100 words to a section already at its word budget), …
- Very specific: line 86
Check my-company/evidence-ledger.json again — evidence may have been added since bind ran.
- Very specific: line 87
If a matching evidence entry exists: attach the ID (<!-- evidence: EV-NNN -->).
- Distinctive: line 88
If no evidence exists: soften or qualify the claim to match what *is* demonstrable, and remove the marker.
- Very specific: line 89
If the claim is central and cannot be softened without gutting the argument: leave the marker, add <!-- PATCH-BLOCKED: CLAIM-UNSUPPORTED — …
- Distinctive: line 90
Never delete a claim without notifying the user in the patch log.
- Distinctive: line 94
Strengths (S): If Gold Team identified a Strength in a section, do not modify that section for any reason other than an explicit D or W …
- Distinctive: line 95
Sections with no findings: Do not edit sections that Gold Team found acceptable and that have no CLAIM-UNSUPPORTED markers.
- Distinctive: line 96
Voice or style: Do not tighten prose, cut marketing language, or improve rhythm.
- Distinctive: line 97
Structure: Do not reorder sections, split or merge paragraphs, or add new sections unless a Deficiency explicitly requires missing content …
- Distinctive: line 98
The narrative spine: Do not change the positioning line or through-line.
- Distinctive: line 104
Before modifying any file, write the current bound draft to a backup:
- Distinctive: line 108
Do this for every file you will modify.
- Very specific: line 150
If any Deficiencies remain unresolved (blocked), do NOT recommend /export-proposal — the draft has a material gap.
- Very specific: line 157
Spine check after patching: After all patches are applied, re-read working/narrative-spine.md and verify the through-line still reads …
≈ “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.
Files
Proposal Patcher Skill
Purpose
Apply the audit patch list — and only the patch list — to the bound drafts.
The bound drafts (drafts/<section>.md) have been written in voice, verified for evidence, and scored by the Gold Team. The audits produced located findings: Weaknesses, Deficiencies, and unsupported-claim flags. This skill resolves exactly those findings — nothing else. The voice and structure installed by the two-pass writer are preserved. The spine's through-line survives.
This is the last content-changing step before /export-proposal.
When to Use
Run after both /evidence-check and /red-team-review --mode=gold have written their findings. Run before /export-proposal.
Track B white-paper workflow:
/narrative-spine → human sign-off
/proposal-writer → draft-loose, then bind
/evidence-check → CLAIM-UNSUPPORTED findings
/red-team-review → Gold Team W/D findings
/proposal-patcher <<< here
/export-proposalDo not run /proposal-editor before this skill. They are mutually exclusive final-pass strategies: the editor does a global pass on structure-first prose; the patcher does surgical fixes on narrative-first prose. Running both defeats the purpose of Track B.
Inputs
Required — will not proceed without these
reviews/gold-team-scorecard.md— must exist and contain at least one Weakness (W) or Deficiency (D). If it exists but has only Strengths, confirm with the user before exiting (the drafts may be submittable as-is).reviews/evidence-check.md— must exist. If stale (older than anydrafts/*.mdfile), warn and recommend re-running/evidence-checkfirst.drafts/<section>.md— the bound drafts produced by/proposal-writer --mode=bind. The loose drafts indrafts/loose/are untouched.
Read if available
working/narrative-spine.md— the through-line the patched drafts must still carry after edits.reviews/technical-review-drafts.md— if/technical-review --phase=draftswas run; include its findings in the patch list.reviews/adversarial/round-<N>-findings.md— if/adversarial-review --mode=single(or--mode=ingest) produced findings without patching them; include the accepted Material/Improvement findings in the patch list (never the Polish appendix). When/adversarial-reviewruns in loop mode it applies its own patches under this skill's rules — don't re-apply the same round here.
Build the Patch List First
Before touching any draft, build a complete patch list. Present it to the user as a table:
| # | Source | Finding Type | Location | Finding Summary | Proposed Fix |
| 1 | evidence-check | CLAIM-UNSUPPORTED | drafts/proposed-approach.md §2 | "... 14 months in disconnected mode ..." | Soften to "demonstrated in disconnected conditions" or attach EV-022 if ledger match found |
| 2 | gold-team W | Weakness | drafts/outcomes-and-value.md §1 | "Outcomes not quantified" | Add a specific metric from working/capability-matrix.md |
| ... |
Then stop and confirm. Ask the user: "This is the patch list — N findings. Shall I apply all, or are there any you want to skip or handle manually?"
This is deliberate. The patch list is cheap to review and expensive to undo. Bill's judgment at this point is higher-value than the agent's — he knows which Gold Team findings reflect real gaps vs. scoring bias.
After confirmation, apply the patches.
Patch Rules
What to patch (in order)
1. Deficiencies (D) — must address
A Deficiency is a material gap that would likely result in an Unacceptable rating if unaddressed. For each D:
- Locate the referenced section and paragraph.
- Make the minimal addition or change that addresses the stated gap.
- Write in the section's existing voice — match cadence, sentence length, register.
- Do not write a new section; add to the section that should carry the point.
- If a Deficiency cannot be addressed without fabricating a capability or past performance, flag it for the user:
<!-- PATCH-BLOCKED: <D-item> — cannot address without unsupported claim. User action required. -->. Do not invent.
2. Weaknesses (W) — address where the fix is clear
A Weakness reduces the rating but does not make the section Unacceptable. For each W:
- Same surgical approach: locate, add minimally, match voice.
- If the fix would require a significant structural change (reordering sections, adding >100 words to a section already at its word budget), flag for the user rather than silently overbuilding:
<!-- PATCH-NOTE: W-item addressed partially — consider full revision if this factor is decision-critical. -->.
3. CLAIM-UNSUPPORTED markers — resolve
For each <!-- evidence: CLAIM-UNSUPPORTED --> found in the bound drafts:
- Check
my-company/evidence-ledger.jsonagain — evidence may have been added since bind ran.- If a matching evidence entry exists: attach the ID (
<!-- evidence: EV-NNN -->). Do not change the claim text. - If no evidence exists: soften or qualify the claim to match what is demonstrable, and remove the marker. Write the softer version in voice — don't just prepend "We believe that..."
- If the claim is central and cannot be softened without gutting the argument: leave the marker, add
<!-- PATCH-BLOCKED: CLAIM-UNSUPPORTED — user must supply evidence or remove claim. -->.
- If a matching evidence entry exists: attach the ID (
- Never delete a claim without notifying the user in the patch log.
What NOT to patch
- Strengths (S). If Gold Team identified a Strength in a section, do not modify that section for any reason other than an explicit D or W finding in the same section. Do not "improve" Strengths — the risk of degrading them outweighs any benefit.
- Sections with no findings. Do not edit sections that Gold Team found acceptable and that have no CLAIM-UNSUPPORTED markers. Zero-finding sections are done.
- Voice or style. Do not tighten prose, cut marketing language, or improve rhythm. That was proposal-editor's job — and in Track B, the loose draft's voice was intentional. Patcher is not an editor.
- Structure. Do not reorder sections, split or merge paragraphs, or add new sections unless a Deficiency explicitly requires missing content that has no existing home.
- The narrative spine. Do not change the positioning line or through-line. If a Gold Team finding seems to require repositioning the whole proposal, flag it for the user — that is a spine revision, not a patch.
Output
1. Pre-patch backup
Before modifying any file, write the current bound draft to a backup:
reviews/pre-patch/<section-id>.mdDo this for every file you will modify. The backup is the restore point if Bill disagrees with a patch.
2. Patched drafts in place
Apply all confirmed patches directly to drafts/<section-id>.md. The patched file is what /export-proposal will read.
3. Patch log
Write reviews/patch-log.md:
# Patch Log — [Proposal Name]
**Generated:** <timestamp>
**Findings reviewed:** <N total>
**Patches applied:** <N applied>
**Blocked / flagged for user:** <N>
## Applied Patches
| # | Source | Type | Location | Change Summary |
|---|---|---|---|---|
| 1 | evidence-check | CLAIM-UNSUPPORTED resolved | drafts/proposed-approach.md L42 | Attached EV-022; claim text unchanged |
| 2 | gold-team W2 | Weakness addressed | drafts/outcomes-and-value.md §1 | Added "reduces operator decision time by ~40% ([Reasoning Product] bench, EV-031)" |
## Blocked / User Action Required
| # | Source | Type | Location | Why Blocked | Required Action |
|---|---|---|---|---|---|
| 3 | gold-team D1 | Deficiency | drafts/team-and-credibility.md | No evidence for claimed ATO experience | User must supply evidence or remove claim |
## Untouched Strengths (preserved)
| Section | Gold Team Strength | Action |
|---|---|---|
| drafts/problem-statement.md | "Compelling problem framing grounded in customer language" | No touch |After Patching
- Tell the user how many findings were resolved, how many were blocked, and list the blocked items prominently.
- Recommend: "Review
reviews/patch-log.mdand resolve any PATCH-BLOCKED items. When ready, run/export-proposal." - If any Deficiencies remain unresolved (blocked), do NOT recommend
/export-proposal— the draft has a material gap. Flag it explicitly.
Discipline Rules
- One finding, one fix. Each patch addresses one finding. Do not use a finding as an excuse to improve nearby prose.
- Minimum effective dose. The right patch is the smallest change that addresses the finding. A one-sentence addition beats a paragraph rewrite.
- The voice stays. Patches must sound like the surrounding prose. If you cannot patch in voice, note it and let the user handle it.
- Spine check after patching. After all patches are applied, re-read
working/narrative-spine.mdand verify the through-line still reads cleanly across the patched drafts. If a patch inadvertently broke the through-line, note it in the patch log.
Activity Trail
On completion, append to working/activity.md:
## <timestamp> — proposal-patcher — <N> patches applied, <M> blocked → reviews/patch-log.mdAppend one JSON line to working/ai-runs.jsonl per reference/schemas/ai-run.schema.json:
{"schema_version":"ai-run.v1","timestamp":"<timestamp>","skill":"proposal-patcher","proposal_id":"<slug>","job_type":"review","provider":"anthropic","model":"claude-opus-4-7","input_tokens_estimate":null,"output_tokens_estimate":null,"cost_estimate_usd":null,"notes":"<N> patches applied, <M> blocked"}Final Rule
The patcher's job is to close the audit's gap list — no more, no less. Every word it doesn't change is a word the writer earned. The voice stays. The spine stays. The Strengths stay.
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
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.

