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

proposal-writer

Skill4 starsChanged 3 months agoMITView on GitHub

Draft evaluator-ready proposal sections from the approved narrative spine and solution architecture, in two passes — draft-loose (confident, in-voice prose) then bind (evidence, verification, compliance). Drafts to the selected narrative operating mode. Writes to drafts/.

29 other projects use a version of this.

Adopted by

Most recently taken up by

Used with

Projects with it also use sop-writer, weekly-report and 5 more.

Other files people use alongside this one.

Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
  • Dokkabei97Skill

    sop-writer

    Creates Standard Operating Procedures for repetitive tasks. Generates step-by-step documentation with checklists, decision trees, and responsibility assignments to eliminate single points of failure.

    Also in 3 projects that use this

  • Microsoft AzureSkill

    weekly-report

    **WORKFLOW SKILL** — Generate weekly executive reports for the azd team. Pulls PR/issue/release data from GitHub, reads changelogs, combines with team notes, and produces a polished markdown report.

    Also in 5 projects that use this

  • akhilkannurSkill

    content-repurposer

    Also in 4 projects that use this

  • claude-office-skillsSkill

    meeting-notes

    >

    Also in 3 projects that use this

  • LangChainSkill

    linkedin-post

    Write a LinkedIn post based on research findings or a given topic. Use this skill when asked to create LinkedIn content, professional posts, or thought leadership pieces.

    Also in 3 projects that use this

  • seikaikyoSkill

    interview-me

    需求不明時的意圖萃取訪談:一次一題、每題附上自己的猜測、聽出「真正想要 vs 覺得應該要」,直到能預測使用者反應(約 95% 信心)才動工。適用:需求缺少對象 / 動機 / 成功標準 / 約束,或使用者點名「訪談我」「先確認一下」「我們確定嗎」。明確自足的指示、純資訊查詢、機械性操作不適用。

    Also in 3 projects that use this

The file

Show its rules and text

We split this skill into the rules it gives an agent

29 rules. 0 are common, 17 are distinctive, and 12 are very specific to this project.

  • 0

    Common

    Most of the 21 edited versions we read keep these.

  • 17

    Distinctive

    This author's own choices. Worth a look.

  • 12

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Read working/proposal-type.md and reference/narrative-operating-modes.md.

    line 83
  2. Distinctive:

    Goal: write each section as a confident argument a senior capture lead would make explaining the case to a sharp colleague — in the …

    line 99
  3. Distinctive:

    Before composing any prose:

    line 105
  4. Distinctive:

    No mandatory eight-element section template, in any order.

    line 115
  5. Distinctive:

    No mechanical four-pattern enforcement.

    line 116
  6. Distinctive:

    No writing "to pass Gold Team."

    line 117
  7. Distinctive:

    When the evaluation model exists, add one output rule for every section: lead with what the listed evaluation factors reward.

    line 132
  8. Distinctive:

    When neither artifact exists, skip this step entirely and draft exactly as described below — this injection changes nothing for proposals …

    line 136
  9. Distinctive:

    Do NOT invent capabilities, past performance, customers, certifications, or numbers.

    line 146
  10. Very specific:

    Do NOT make any claim working/capture-intent.md lists under "Proposal Must Avoid" / prohibited claims.

    line 147
  11. Very specific:

    If you need a fact you do not have, write the sentence and leave a [NEEDS: <what>] marker inline rather than inventing it.

    line 148
  12. Very specific:

    Check every factual claim against working/architecture-concept.md, working/solution-strategy.md, and my-company/evidence-ledger.json (when …

    line 162
  13. Very specific:

    Read reference/proposal-writing-patterns.md and the type's pattern applicability table.

    line 171
  14. Distinctive:

    Do NOT reorder the loose draft to match a fixed element sequence.

    line 175
  15. Distinctive:

    Do NOT add a pattern to check a box.

    line 176
  16. Very specific:

    Ghosting: only ghost competitor weaknesses documented in working/competitor-assessment.md.

    line 180
  17. Very specific:

    Read reference/preventable-gold-team-findings.md and run its five-check sweep on each bound section.

    line 185
  18. Very specific:

    Add the missing enumeration minimally, in the section's own voice.

    line 215
  19. Very specific:

    If working/compliance-matrix.md exists, every section bound must update it — this is how the writer hands off to compliance-check and …

    line 231
  20. Very specific:

    If a requirement has no obvious home, do NOT assign it to an unrelated section — leave it Planned and flag it for compliance-check.

    line 236
  21. Very specific:

    Do not flip rows to Covered — that is compliance-check's call, or a human's.

    line 237
  22. Distinctive:

    Never delete rows, never edit the Requirement text column, never renumber Req IDs.

    line 238
  23. Very specific:

    After binding, recommend the user run /compliance-check.

    line 240
  24. Distinctive:

    No trailing explainer clauses: ", which is...", ", which means...", ", which allows...", "which is what makes...".

    line 276
  25. Distinctive:

    No pseudo-clefts or abstract-subject openers: "What does not vary is...", "The value to this effort is...", "The consequence for the Army …

    line 277
  26. Distinctive:

    Do not explain the solicitation back to the evaluator ("which is the level this Challenge targets", "this is why the Challenge requires").

    line 278
  27. Distinctive:

    Prefer verbs over nominalizations: "the Army deploys" over "deployment occurs".

    line 280
  28. Distinctive:

    Never use internal product-component names (Garrison, Operator, ODIN, DORIS).

    line 282
  29. Distinctive:

    Default narrative hierarchy: military-native models, then the secure runtime that executes them, then Army-specific adaptation, then …

    line 283

“Common” is measured: kept by at least 6 in 10 of the 21 edited versions of this file we could read. “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

Proposal Writer Skill

Purpose

Turn the approved narrative spine and solution architecture into proposal-ready section drafts saved to local files.

The Two-Pass Model

The writer runs in two passes, because composing a compelling argument and verifying every fact are different cognitive jobs — and doing them at once produces stilted, slot-filled prose. The first pass writes the argument; the second pass makes it traceable.

ModeWhat it doesReadsWrites
draft-loose (Pass 1)Writes each section as a confident, in-voice argument from the narrative spine, in the selected narrative operating mode. Evidence markers, the section completeness checklist, pattern enforcement, and compliance-matrix updates are suspended.spine, storyboard, capture-intent, architecture, voice guide, operating modesdrafts/loose/<section-id>.md
bind (Pass 2)Takes the loose draft and makes it submission-ready: attaches evidence citations, verifies every claim, flags unsupported claims, adds any genuinely missing scoring element, updates the compliance matrix. Touches prose only for verification — never for style.the loose draft + evidence ledger, compliance matrix, proposal-plan, competitor-assessmentdrafts/<section-id>.md
full (default)Runs draft-loose then bind for each section, in sequence.—both

Invocation. /proposal-writer with no mode runs full. /proposal-writer --mode=draft-loose or --mode=bind runs a single pass — useful for inspecting the loose draft before binding, or re-binding after a hand edit.

Why the split. When generation has to satisfy the compliance matrix, the pattern checklist, the section template, and the evidence protocol while composing, there is no room left to invent rhythm — the result reads like filled-in slots. draft-loose gives composition full freedom; bind does verification afterward, on real prose, without touching cadence.

Note: this skill has two senses of "mode." The pass mode (draft-loose / bind / full) selects which pass runs. The narrative operating mode (compressed-brief, solution-brief, etc., below) is the prose strategy for the response. Every draft-loose run is written in one narrative operating mode.

Inputs

Always read (mandatory)

  1. working/proposal-type.md — read first. If this skill is in skipped_skills, exit with "Skipped for type <type_id>." Adapt tone, section set, and page targets to evaluator_framing + page_target + section_patterns.
  2. working/narrative-spine.md — read second, as primary orientation. The approved one-page argument the whole proposal makes (produced by /narrative-spine). Every section drafted advances one movement of this argument. Do not invent a new argument — draft the one the spine commits to. If working/narrative-spine.md does not exist, recommend the user run /narrative-spine first; proceed only if the storyboard is strong enough to draft from on its own.
  3. working/proposal-plan.md — eval criteria and win themes.
  4. working/solution-strategy.md
  5. working/architecture-concept.md
  6. reference/section-patterns/<section_patterns>.md — derived from section_patterns in working/proposal-type.md. Declares section order, required/optional sections, per-section purpose.
  7. reference/editorial-voice-guide.md — voice discipline applies in both passes. Loose is free in form and confidence; it is not free to be sloppy or to use marketing language. In particular, even the loose pass must follow the guide's "Dashes, Self-Narration, and Prohibited Claims" rules: no dashes as sentence punctuation (em-/en-dash or -- — this is a HIGH prose-lint finding that blocks export), no self-narration of the proposal's own honesty, and no FedRAMP/ATO/"accredited"-style claims without an evidence-ledger cite.
  8. reference/voice-anchors/*.md — voice cadence exemplars. Read these BEFORE composing the loose draft. Imitate their rhythm and clinical confidence; do NOT reuse their content. Anchors take precedence over a banned-words list when the two conflict. (Canonical source: reference/PROSE-QUALITY-DOCTRINE.md.)
  9. reference/narrative-operating-modes.md — the prose strategy for the response. Select the operating mode before draft-loose begins (see "Narrative Operating Mode" below).
  10. working/compliance-matrix.md — the traceability map. Required reading for bind (see "Compliance Matrix Maintenance"); not needed for draft-loose.
  11. reference/preventable-gold-team-findings.md — the bind-pass completeness sweep (Pass 2, Step 2b) that closes the preventable ~70% of Gold-Team findings before review. Required reading for bind; not a draft-loose input (loose stays free of Gold-Team constraints by design).

Read if relevant

  • working/storyboard.md — if /proposal-storyboard was run; the section-by-section decomposition of the spine. Primary draft-loose planning input when present. Carries the per-section Reader Movement, Narrative Mode, Transition Job, and Compression Rule fields — plus the factor-keyed fields (Target Evaluation Factors, Evaluated Strength to Earn, Discriminator) consumed by the "Write toward the rubric" step below.
  • working/evaluation-model.md — if /proposal-manager Step 4b was run; this solicitation's typed rubric (factors, weighting, pass/fail gates, constraints with provenance). Read the "Drafting Context Block" section. When present, these criteria override generic doctrine about what evaluators reward.
  • working/capture-intent.md — strategic guidance: customer beliefs, prohibited claims, posture, ghosting direction. Binding on both passes — draft-loose may not make claims capture-intent prohibits.
  • working/requirement-matrix.md, working/capability-matrix.md, working/assumptions-and-risks.md — grounding for the relevant sections.
  • my-company/evidence-ledger.json — required for bind when Phase C is enabled (evidence citation).
  • working/competitor-assessment.md — for ghosting; only ghost weaknesses documented here.
  • working/graphics-brief.md — for action captions in bind.
  • reference/proposal-conventions/<vehicle-id>.md — calibrated conventions, when a convention file exists. Mapping: full-proposal→far-rfp.md; sbir→sbir.md; gsa-mas-task-order→gsa-mas.md (plus a Security Volume per reference/section-patterns/security-volume.md when cleared work is required); white-paper→white-paper.md; other patterns fall through to section-patterns alone.

Outputs

Output filenames derive from the pattern file's section_order, not hardcoded. Reading reference/section-patterns/<section_patterns>.md yields a list of section ids; the writer produces one file per required section (and each optional section where source material exists).

  • draft-loose writes drafts/loose/<section-id>.md
  • bind reads drafts/loose/<section-id>.md and writes drafts/<section-id>.md

The loose draft is kept, not overwritten — it is the inspectable record of the argument before verification, and the input bind re-reads.

Dispatch by section_patterns:

section_patternsTypical draft files produced
full-proposalexecutive-summary.md, problem-statement.md, capability-overview.md, technical-approach.md, management-approach.md, past-performance.md, path-forward.md, conclusion.md
white-paperexecutive-summary.md, problem-statement.md, proposed-approach.md, outcomes-and-value.md, team-and-credibility.md (optional), rom-or-next-steps.md (optional)
baaexecutive-summary.md, research-hypothesis.md, technical-approach.md, work-plan-and-milestones.md, principal-investigator-bio.md, prior-research-results.md (optional), schedule-and-gates.md
otaexecutive-summary.md, statement-of-objectives-response.md, prototype-project-scope.md, technical-approach.md, milestone-schedule.md, team-composition.md, data-rights-assertions.md, follow-on-production.md
sbiridentification-and-significance.md, phase-i-feasibility-or-phase-ii-prototype.md, technical-objectives.md, work-plan.md, key-personnel.md, commercialization-plan.md, budget-narrative.md
rficover-letter.md, question-by-question-response.md, capability-summary.md
sources-soughtcompany-identification.md, relevant-experience-table.md, capability-summary.md
romrom.md (single-file artifact, with section_order's subsections as H2 headings within)

Rule: Do NOT produce drafts for sections not listed in the pattern's section_order. If the solicitation calls for a section not in the current pattern set, either (a) add it to the pattern file (update the registry too) or (b) note it as an override in the proposal-brief.

Narrative Operating Mode

Before draft-loose begins, select one narrative operating mode for the response. The mode is the prose strategy — it controls how much setup, depth, and explanation the writing can afford. It is the difference between a draft that fits its proposal type and one the editor has to rescue.

  1. Read working/proposal-type.md and reference/narrative-operating-modes.md. If /proposal-storyboard recorded a mode in its header, use that; the storyboard and the writer must agree.
  2. Select one mode for the response from page_target, submission_mechanism, and evaluator_framing:
    • compressed-brief — 1–5 page briefs, short white papers, tight web-form fields
    • solution-brief — 5–10 page CSO / OTA / white-paper responses
    • scored-volume — FAR, IDIQ, OTA full, and CSO full proposals
    • technical-merit — SBIR and BAA responses
    • market-research — RFI and sources-sought
    • pricing-artifact — ROM and price narratives
  3. Apply the mode from the first sentence — to paragraph length, amount of setup, technical depth, and use of bullets. Do not leave length or register correction for the editor.

Mode discipline. A two-page white paper does not get a full conceptual tutorial. A scored FAR volume does not get airy thought-leadership prose. An RFI does not get win-theme theatrics. A ROM does not get a sales arc. When in doubt, choose the shorter, plainer mode — overexplaining is the default failure.


Pass 1 — Draft-Loose

Goal: write each section as a confident argument a senior capture lead would make explaining the case to a sharp colleague — in the selected narrative operating mode. Make the case; sound like a person.

Step 0 — Build the response skeleton (rubric-scored types)

For rubric-scored technical types (ota-*, cso-*, far-rfp, idiq-to, sbir-*, baa), the draft is written INTO the customer's own structure, not into an essay shape of the writer's choosing. Calibration evidence (DIA DMA submission diff, 2026-07-26): only 14% of an essay-shaped loose draft survived to submission; the 71% rewrite was substantially a shape problem — the human end-state was organized in the customer's vocabulary (LOE names, factor names, required-content headings), so well-written sentences in the wrong structure died anyway.

Before composing any prose:

  1. From the solicitation (inputs/00_priority/), working/submission-summary.md, and working/evaluation-model.md, extract the customer's own structural vocabulary: named lines of effort / task areas, evaluation factor names, and the required-content list (deliverables, schedule, payment, data rights, POC, past performance).
  2. Emit a response skeleton: the section/heading tree the customer's documents imply, using the customer's terms verbatim (e.g., "Line of Effort 1: <customer's own LOE title>", the factor names as sub-blocks of the technical approach). Reconcile with section_patterns — the pattern file gives the file set; the skeleton gives the headings inside those files.
  3. The voice draft fills the skeleton. The skeleton is never derived from the draft.

For narrative-first types (white-paper, rfi, sources-sought) skip this step — the essay shape is the point there.

Suspended in this pass — do not do any of these now, they belong to bind:

  • No <!-- evidence: --> markers.
  • No compliance-matrix updates.
  • No mandatory eight-element section template, in any order.
  • No mechanical four-pattern enforcement.
  • No writing "to pass Gold Team."

Do:

  • Write toward the rubric (evaluator-upstream injection — additive only). Before composing each section, assemble its drafting context in this order, including each block only when its artifact exists:

    1. Storyboard block — from the section's working/storyboard.md entry: the target evaluation factors to win, the evaluated strength to earn, the acceptance measure (the verifiable commitment the section's prose must state — emit it, don't paraphrase it away), the discriminator, the proof points, and the planned beat order. The draft fills this plan.
    2. Evaluation-model block — the "Drafting Context Block" from working/evaluation-model.md: this solicitation's factors, weighting, pass/fail gates, and constraints. These override generic doctrine when they conflict.

    When the evaluation model exists, add one output rule for every section: lead with what the listed evaluation factors reward; make discriminators and benefits explicit so the evaluator can score them without inference.

    When neither artifact exists, skip this step entirely and draft exactly as described below — this injection changes nothing for proposals without a rubric or storyboard.

  • Draft the spine's argument, in the operating mode. Each section advances one movement of working/narrative-spine.md (the storyboard, if present, says which), at the depth and length the operating mode allows.
  • Move the reader. Each section should move the evaluator from one belief to the next — problem recognized → answer understood → feasibility believed → risk reduced → next action obvious. Use the storyboard's Reader Movement and Transition Job fields when present; infer them when not. A section that only "tells the reader more about our capability" is not earning its place.
  • Open concern-first. Lead with the customer's mission problem, the evaluator's question, a load-bearing number, or a blunt statement of stakes — whatever states the section's point fastest. Do not open with a dramatized or cinematic scene ("Picture the demonstration room...", "In August, four personas will..."); staged openings read as cheesy and weaken the impression that the team understands the mission. Do not open every section with the same theme-statement formula. Vary openings across sections — the human element comes from visible command of the problem, not theatrics.
  • Make transitions carry meaning. Connect sections by consequence, selection, proof, or decision — not by announcing structure ("This section discusses...", "The following paragraphs describe...").
  • Write in voice. Apply reference/editorial-voice-guide.md: direct, clinical, operationally credible; no marketing adjectives; observable facts over claims. Loose means free in form and confidence — not free to be sloppy.
  • Assert with confidence. You may state claims that will need evidence — bind will check them and attach citations or flag gaps. Write the strong version now; do not pre-hedge every sentence.

Honesty boundary — loose is not a license to fabricate.

  • Do NOT invent capabilities, past performance, customers, certifications, or numbers. Every fact must trace to the architecture, solution strategy, or source material.
  • Do NOT make any claim working/capture-intent.md lists under "Proposal Must Avoid" / prohibited claims. Form is free; prohibited claims are still prohibited.
  • If you need a fact you do not have, write the sentence and leave a [NEEDS: <what>] marker inline rather than inventing it.

Output: drafts/loose/<section-id>.md. Clean prose — no evidence markers, no matrix notes.


Pass 2 — Bind

Goal: make the loose draft submission-ready and fully traceable without rewriting it for style. Bind preserves the loose draft's voice and structure. It touches prose only to attach evidence, soften an unsupported or risky claim, or add a genuinely missing scoring element.

For each loose draft, in order:

1. Verify and cite every claim

Check every factual claim against working/architecture-concept.md, working/solution-strategy.md, and my-company/evidence-ledger.json (when Phase C is enabled).

  • For a claim with backing evidence, attach an HTML-comment citation: Our platform operated at [customer] for 14 months in disconnected mode. <!-- evidence: EV-022 -->. HTML comments are invisible in markdown render and docx export but machine-readable. Multiple IDs allowed: <!-- evidence: EV-022, EV-055 -->.
  • For a claim with no ledger backing, mark it explicitly: <!-- evidence: CLAIM-UNSUPPORTED -->. Do NOT delete the claim and do NOT silently launder it into confident prose. If the claim is also strategically risky, soften, localize, or qualify it — then keep the marker.
  • Resolve every [NEEDS: ...] marker from Pass 1 — supply the fact if it exists, or convert to CLAIM-UNSUPPORTED and flag for the user.
  • Precedence for citation sources: my-company/evidence-ledger.json > drafts/past-performance.md > inputs/02_yourCompany/past-performance.md. See reference/schemas/evidence-ledger.schema.json.

2. Check — do not enforce — the winning patterns

Read reference/proposal-writing-patterns.md and the type's pattern applicability table. The four patterns (theme statement, discriminator proof point, action caption, ghosting) are a completeness check on the loose draft, not a generation template.

For each scoring section, confirm it lands: a clear point up front, at least one discriminator proof point, a tie to the evaluation criterion / requirement, team credibility, and an evaluator takeaway. If one is genuinely missing and adding it strengthens the section, add it — minimally, in the section's own voice, where the argument naturally wants it.

  • Do NOT reorder the loose draft to match a fixed element sequence.
  • Do NOT add a pattern to check a box. A forced theme statement on a 200-word boilerplate section makes it worse.
  • Respect the operating mode. In compressed-brief and solution-brief modes, do not add every pattern element — the shortest sequence that lands the movement (claim, mechanism, proof, value, next action) is correct, and an element that makes the prose feel procedural should be folded into the sentence that already carries the point. scored-volume sections carry the full set; market-research and pricing-artifact carry almost none.
  • Patterns apply to scoring sections (technical approach, management, past performance, differentiators, risk). Reps and certs, CDRL lists, staffing rosters get only what the solicitation requires.
  • White papers: apply theme statement and discriminator proof point where a section makes a scoring claim; skip the discriminator/credibility/ghosting sequence in informational sections. Self-positioning stays in the dedicated position section, never in evidence tables.
  • Ghosting: only ghost competitor weaknesses documented in working/competitor-assessment.md. If it is absent, do not invent ghosts.
  • Action captions: for each referenced graphic, read its brief in working/graphics-brief.md and write a caption that asserts what the graphic proves.

2b. Sweep the preventable Gold-Team findings

Read reference/preventable-gold-team-findings.md and run its five-check sweep on each bound section. These are the "fill-the-field / show-your-work" failures the Gold Team otherwise catches late; closing them now raises the floor so the adversarial pass spends its judgment on the residue that needs an evaluator's frame. This is a completeness sweep on real prose, not a generation template — same discipline as Step 2.

  1. A scored or required field is never a placeholder. If a required datum (ROM/price, NAICS, contract number, POC, period of performance, a required certification) is missing, surface it to the user as a blocker — convert the [NEEDS: …] to a flagged blocker; never leave a bracketed blank or [TBD] in the body. An unscorable required field caps pWin and can read as non-responsive. (Stricter than the White Glove placeholder scan, which runs too late to protect the score.)
  2. Name key personnel with a one-line relevant credential where the type expects named personnel; a role-only "experienced team" is a high-impact finding. If a name is genuinely unassigned, carry [NEEDS: named <role>] as a blocker.
  3. Define each acronym / term of art on first use; one canonical term per concept across sections (no double-labeling).
  4. Name the specific comparator / benchmark result, not "state-of-the-art" or "leading." If the specific number isn't held, it is a CLAIM-UNSUPPORTED (Step 1), not a vague superlative.
  5. Preempt the evaluator's obvious unasked question for each scoring section (impact level, who validates, how it is tested, transition path) rather than leaving it for the reviewer to raise.

Apply per the type table in the reference. Do not add a check the type marks Not applicable (an RFI gets no personnel sweep; a ROM gets no comparator sweep).

2c. Operational-specificity sweep

Rule of thumb (from submission-diff calibration): a capability sentence that contains no noun-list and no number will be rewritten by a human before submission. The enumerations that give a winning response its texture almost always already exist in working/ — the failure mode is suspending them for voice and never reinstalling them.

For each capability claim in a scoring section, check that the bound prose carries its operational specifics, pulled from the artifact that already holds them:

Claim typeRequired specificitySource
An interface / APIThe surface enumerated (endpoints, operations, metadata)working/architecture-concept.md
A support package / deliverableThe artifact inventory named (e.g., SSP inputs, SBOM, POA&M inputs, ConMon plan)architecture-concept, requirement-matrix
Monitoring / metricsThe metric list (who, what, by model/workflow/user)architecture-concept, storyboard
Evaluation / TEVVThe measure list and who adjudicatesstoryboard, evaluation-model
A process (update, onboarding, adaptation)The step sequence + cycle time + rollback/failure patharchitecture-concept
Schedule / scopeMilestone table with dates/months; counts ("two to three models")prototype scope, pricing-inputs

Add the missing enumeration minimally, in the section's own voice. If the specifics genuinely don't exist anywhere in working/, that is an architecture gap — flag it to the user rather than inventing detail.

2d. Constrained-claim sweep (overclaim-by-scope)

The claim form that survives every review names its own boundary. For each product-capability claim, verify the draft states the scope limit in-line, using the constrained idioms:

  • supports / contributes to X, not performs X (e.g., "contributes the factual-support measure to TEVV rather than performing TEVV itself")
  • initiates / supports accreditation, not achieves ATO
  • capability applies where its precondition holds ("federated where its endpoint is available")
  • component is core but separable (adds capability without becoming a lock-in or failure point)
  • what cannot be verified is labeled as such, not silently included

An unconstrained capability claim is a latent Gold-Team weakness even when true — the evaluator reads the strongest possible interpretation and scores the risk of it. Write the boundary before a reviewer asks for it.

3. Update the compliance matrix

If working/compliance-matrix.md exists, every section bound must update it — this is how the writer hands off to compliance-check and red-team.

For each section:

  1. Identify which Req IDs the section addresses (find rows whose Requirement text is treated in this section).
  2. For each matching row set: Section (draft filename + heading path, e.g. drafts/technical-approach.md Section 3.2); Page (TBD until the Word doc is assembled); Status (Drafted for substantive coverage, Partial for light coverage needing more, leave Planned if explicitly deferred); Evidence (a 1-line pointer).
  3. If a requirement has no obvious home, do NOT assign it to an unrelated section — leave it Planned and flag it for compliance-check.
  4. Do not flip rows to Covered — that is compliance-check's call, or a human's.
  5. Never delete rows, never edit the Requirement text column, never renumber Req IDs.

After binding, recommend the user run /compliance-check.


Final Pass (after bind or full)

  • Redundancy check — the same point should not be argued in two sections.
  • Narrative-mode check — the prose matches the selected operating mode; no over-explaining for the type, no under-explaining.
  • Reader-movement check — each section moves the evaluator; transitions carry consequence, selection, proof, or decision.
  • Unsupported-claim check — every CLAIM-UNSUPPORTED marker is intentional and flagged, not an oversight.
  • Requirement coverage check — via the compliance matrix.
  • Spine check — every section still advances its movement of the narrative spine; the through-line survives.

Critical Rule

All draft content must be written to files (drafts/loose/ and drafts/) — not just displayed in chat.

Activity Trail

On completion, append to working/activity.md:

## <timestamp> — proposal-writer [<mode>] — drafted/bound <N sections>, <M matrix rows updated> → drafts/

Append one JSON line per pass to working/ai-runs.jsonl per reference/schemas/ai-run.schema.json with job_type: "drafting" and notes recording the pass mode and the narrative operating mode.

Voice Standard: Active, Declarative, Operator-Focused (adopted 2026-07-30)

Adopted from the army-brevity-companion voice retrospective. The failure this standard prevents is not passive voice; it is prose built around explanation, qualification, and self-justification: sentences that prove the writer understood the requirement instead of stating what the solution does. scripts/prose-lint.py flags the mechanical signatures as ADVISORY (ai_proposalese terms, explainer_tail pattern).

Rules:

  1. Lead each sentence with an actor: [Your Company], the Companion, the Army, staff, the Government. Put the verb in the first half.
  2. One main idea per sentence. Split any sentence carrying claim + explanation + implication + solicitation mapping.
  3. State the capability, then the evidence. Do not narrate the reasoning that connects them.
  4. No trailing explainer clauses: ", which is...", ", which means...", ", which allows...", "which is what makes...". Promote the point to its own sentence with an actor, or fold it into the verb.
  5. No pseudo-clefts or abstract-subject openers: "What does not vary is...", "The value to this effort is...", "The consequence for the Army is...", "The result is a...". Write "The Army can...", "[Your Company] delivers...".
  6. Do not explain the solicitation back to the evaluator ("which is the level this Challenge targets", "this is why the Challenge requires"). State that the solution meets the requirement and move on.
  7. Ration contrast constructions ("not X but Y", "rather than", "does not merely"). Keep only the ones carrying a load-bearing distinction; more than a handful reads as a debate brief.
  8. Prefer verbs over nominalizations: "the Army deploys" over "deployment occurs"; "staff produce" over "the resulting communication".
  9. Vary sentence length; break the balanced-triad cadence deliberately. Target a 15 to 22 word average.
  10. Never use internal product-component names (Garrison, Operator, ODIN, DORIS). [Your Company] is the integrated product; name [Runtime Product] and Ground Truth only where they help the evaluator.
  11. Default narrative hierarchy: military-native models, then the secure runtime that executes them, then Army-specific adaptation, then verification and traceability, then workflow integration, then measured improvement.

Test for every paragraph: name its single purpose, make the first sentence state that purpose with an actor and a verb, and let the remaining sentences carry only evidence or execution detail. Write as though a staff officer has 30 seconds.

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.