evidence-check
Skill4 starsChanged 3 months agoMITView on GitHub
Audit draft evidence citations against my-company/evidence-ledger.json — flag CLAIM-UNSUPPORTED, typos, retired/restricted IDs. Run after proposal-writer, before red-team-review --mode=gold.
6 other projects use a version of this.
Adopted by
Most recently taken up by
- Bhenderlong/vallorix-mcp-plugintheir own version, first seen Nov 8, 2026
- ashe-li/agent-skillstheir own version, first seen Oct 29, 2026
- jerseycheese/agent-skillstheir own version, first seen Oct 28, 2026
- MichaelRDionne/clinical-agent-skillstheir own version, first seen Oct 8, 2026
- Kaagemusha/agent-skillstheir own version, first seen Sep 27, 2026
- reutavidan/zetesis-plugintheir own version, first seen Aug 6, 2026
Other versions6 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 6
- jerseycheese/agent-skills★ 4same name
- ashe-li/agent-skills★ 2same name
- Bhenderlong/vallorix-mcp-plugin★ 0same name
- Kaagemusha/agent-skills★ 0same name
- MichaelRDionne/clinical-agent-skills★ 0same name
- reutavidan/zetesis-plugin★ 0same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
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
- export-proposalSkill
- import-from-captureSkill
- narrative-spineSkill
- new-proposalSkill
- opportunity-quick-lookSkill
- past-performanceSkill
- pricing-analystSkill
- proposal-editorSkill
- proposal-graphicsSkill
- proposal-managerSkill
- proposal-patcherSkill
- proposal-solution-architectSkill
- proposal-storyboardSkill
- proposal-writerSkill
- red-team-reviewSkill
- setup-companySkill
- statusSkill
- submission-summarySkill
- technical-reviewSkill
The file
We split this skill into the rules it gives an agent
5 rules. 0 are common, 1 is distinctive, and 4 are very specific to this project.
0
Common
Most of the 5 edited versions we read keep these.
1
Distinctive
This author's own choices. Worth a look.
4
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 62
If the ledger file does not exist, emit a warning: "No evidence ledger found at my-company/evidence-ledger.json.
- Very specific: line 76
Not in any ledger set → UNKNOWN_ID (typo or missing entry)
- Very specific: line 87
Update working/compliance-matrix.json's summary block with an evidence_coverage_pct field (writer of that file is compliance-check.
- Distinctive: line 170
Never invent evidence IDs: If an unknown ID is found, flag it.
- Very specific: line 175
If my-company/evidence-ledger.json is missing, emit guidance rather than failing:
“Common” is measured: kept by at least 6 in 10 of the 5 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.
Files
/evidence-check
Purpose
Phase C of v1.5 introduces evidence-grounded drafting: every major claim in a proposal links to an evidence item in my-company/evidence-ledger.json via an HTML comment marker. This skill audits those links.
Running this skill tells you:
- Unsupported claims — the writer flagged a claim but couldn't find evidence (or a claim slipped in without a marker at all — detection is heuristic)
- Typo'd or stale IDs — the draft cites
EV-022but the ledger has no such entry (or it's retired) - Restricted-context violations — an
approval_status: restricteditem was cited in a proposal type the restrictions forbid - Unused evidence — approved items in the ledger that no draft cites — strong proof points you might be missing in the narrative
Gold Team in /red-team-review automatically converts CLAIM-UNSUPPORTED markers into Weakness findings, so running /evidence-check and resolving its output before Gold Team saves you rework.
Citation format (established in proposal-writing-patterns.md)
Supported claim — HTML comment with one or more evidence IDs
Our platform has operated at [customer] for 14 months in fully disconnected mode. <!-- evidence: EV-022 -->Multiple IDs allowed:
On-device inference <!-- evidence: EV-001 --> has been validated at operational scale <!-- evidence: EV-022, EV-055 -->.Unsupported claim — explicit marker
Our on-device inference is the fastest in the industry. <!-- evidence: CLAIM-UNSUPPORTED -->Writer should mark these explicitly when it can't find evidence. Gold Team treats them as Weaknesses until resolved.
What it reads
working/proposal-type.md— determine proposal-type for restriction checksmy-company/evidence-ledger.json— the approved evidence librarydrafts/*.md— scan for<!-- evidence: ... -->markersworking/compliance-matrix.json(if exists) — read rows to understand which claims should be evidence-backedworking/proposal-plan.json(if exists) — read discriminators (each should have evidence_refs)
Algorithm
Step 1: Parse ledger
Load my-company/evidence-ledger.json. Build an index:
approved_ids: set of IDs withapproval_status == "approved"retired_ids: set withapproval_status == "retired"restricted_ids: set withapproval_status == "restricted"+ restrictions textpending_ids: set withapproval_status == "pending_review"unknown_ids: for tracking typos — populated as we scan drafts
If the ledger file does not exist, emit a warning: "No evidence ledger found at my-company/evidence-ledger.json. Run the Phase C seeding process (see reference/examples/evidence-ledger.example.json) to create one." Exit with a "framework-not-ready" status but don't fail the skill.
Step 2: Scan drafts for citations
For each drafts/*.md file, extract all <!-- evidence: ... --> markers. Parse the comma-separated IDs inside. Record:
cited_ids: set of all IDs cited anywhere in draftscitation_locations: mapid → [ (file, line, surrounding_sentence) ]unsupported_claims: list of(file, line, sentence)for eachCLAIM-UNSUPPORTEDmarker
Step 3: Classify citations
For each cited ID:
- In
approved_idsand no restriction conflict → OK - In
retired_ids→ RETIRED finding (remove or replace) - In
restricted_idsand restrictions forbid this proposal type → RESTRICTED_VIOLATION - In
pending_ids→ PENDING_REVIEW warning (don't submit until approved) - Not in any ledger set → UNKNOWN_ID (typo or missing entry)
Step 4: Find unused evidence
approved_ids - cited_ids = evidence the user has but didn't use. For each, note the relevance_tags — if they overlap with the current proposal's domain (from working/proposal-plan.json win themes or the proposal type's evaluator framing), surface as "Consider citing."
Step 5: Compute coverage metric
total_claim_markers= count of all<!-- evidence: ... -->markers across draftssupported= markers with at least one approved, non-violated IDunsupported= markers withCLAIM-UNSUPPORTEDor only retired/unknown IDscoverage_pct= supported / total_claim_markers × 100
Update working/compliance-matrix.json's summary block with an evidence_coverage_pct field (writer of that file is compliance-check; this skill is allowed to update the evidence_coverage_pct field only).
Or if the matrix doesn't exist (e.g., type has no compliance_sources), just report coverage inline in the review.
What it writes
reviews/evidence-check.md
# Evidence Check — <proposal name>
**Date:** <timestamp>
**Ledger:** my-company/evidence-ledger.json (N approved, M pending, K retired, J restricted)
**Drafts scanned:** N files
**Total claim markers:** N
**Supported:** N (coverage: X%)
**Unsupported:** N
**Unknown IDs:** N
**Retired IDs cited:** N
**Restricted violations:** N
## Unsupported claims
(Each flagged `CLAIM-UNSUPPORTED` marker)
- **drafts/technical-approach.md §3.2 line 47:** "Our on-device inference is the fastest in the industry."
- Recommendation: find/add evidence (consider EV-055 which benchmarks pipeline speed) OR soften claim OR remove.
## Unknown / typo'd IDs
- **drafts/executive-summary.md line 18:** cited `EV-222` — not found in ledger. Likely typo for `EV-022`?
## Retired IDs cited
(list — these must be replaced or removed before submission)
## Restricted violations
(list — these citations violate the evidence item's restrictions for the current proposal type)
## Unused approved evidence worth considering
- **EV-031 (named_personnel):** Dr. Jane Doe — DARPA continuity. Tags: darpa, neuro-symbolic. Fits this proposal's DARPA emphasis.
- **EV-067 (capability_claim):** FedRAMP Moderate. Tags: fedramp, il5, ato. Might strengthen AoI-4 response.
## Coverage summary
- Total claim markers: N
- Supported: N (X%)
- Unsupported: N
- Target: zero unsupported markers before Gold TeamUpdated working/compliance-matrix.json (if present)
Adds or updates summary.evidence_coverage_pct.
Activity trail
Append to working/activity.md:
## <timestamp> — evidence-check — <N supported> / <M unsupported> / <K unknown> / <J retired>, coverage <X>% → reviews/evidence-check.mdAppend to working/ai-runs.jsonl:
{"schema_version":"ai-run.v1","timestamp":"<ts>","skill":"evidence-check","proposal_id":"<slug>","job_type":"review","notes":"deterministic scan, no AI invocation"}Note: this skill is primarily deterministic — it's a structural diff, not an AI inference task. It should log a run entry anyway for audit completeness, but model, input_tokens_estimate, and output_tokens_estimate can all be null.
When to run
- After every
/proposal-writerpass — catches typos and unsupported claims while the writer's context is fresh - Before every
/red-team-review --mode=gold— Gold Team treats unsupported claims as automatic weaknesses, so fix them first - Before final
/export-proposal— submission gate: target zero unsupported, zero retired, zero restricted violations
Rules
- Never modify drafts/*.md. This skill is read-only over drafts. User resolves findings by editing drafts themselves or re-running writer on specific sections.
- Never modify the evidence ledger. Ledger is curated by the user or
/setup-company. This skill reports; user curates. - Never invent evidence IDs. If an unknown ID is found, flag it. Do not "fix" it by creating a new ledger entry.
- Always update
working/compliance-matrix.jsonwith the evidence_coverage metric (if the matrix exists) so Gold Team and dashboard see current coverage.
When the ledger doesn't exist yet
If my-company/evidence-ledger.json is missing, emit guidance rather than failing:
No evidence ledger found.
This workspace expects
my-company/evidence-ledger.jsonto be populated from your actual past performance, capabilities, and credentials. Seereference/examples/evidence-ledger.example.jsonfor the shape, andreference/schemas/evidence-ledger.schema.jsonfor the full schema.To seed your ledger from your existing
my-company/content:
- Read
my-company/past-performance.mdandmy-company/capabilities.md- Create one
EV-###entry per distinct proof point (contract, deployment, credential, capability claim)- Set
approval_status: "approved"for items cleared for public citation- Commit the file locally (it's gitignored — stays on your machine)
Until the ledger exists, Phase C drafts can still be produced; they just won't be citation-backed. Plan to seed the ledger before your next major proposal.
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.

