source-locked-verification
No Inference / Source-Locked Verification. Forces Claude to answer ONLY from user-provided materials and/or online sources actually accessed — no inference, no assumptions, no gap-filling. Every factual, legal, numerical, or procedural claim must be anchored to a cited source. Use whenever Claude reviews documents, summarises evidence, checks accuracy, drafts submissions, creates timelines, extracts facts, checks citations, analyses rules, prepares legal arguments, compares documents, verifies claims, produces chronologies, works from uploaded materials, performs legal research, checks case st
npx skills add lawve-ai/awesome-legal-skills --skill source-locked-verification-larissa-meredith-flister --agent claude-code
Same command for any agent — swap --agent for codex, cursor, copilot.
Weekly change comes from our own snapshots, not the repository page — it measures attention, not adoption.
What it does
For tasks like reviewing documents, summarising evidence, drafting submissions, creating timelines, extracting facts, checking citations, and preparing legal arguments, this skill forces the agent to answer only from user-provided materials and/or online sources actually accessed. It requires that every factual, legal, numerical, or procedural claim be anchored to a cited source and avoids inference or gap-filling.
How it works
The agent must operate in source-locked mode: rely solely on materials provided by the user or online sources actually accessed during the task. When online research is used, citations must come from verified sources. The agent must not assert details unless they are expressly stated in the materials or verified online sources. The framework provides rules for online necessity, evidence anchoring, source hierarchy, output categorization, and handling of missing information, with explicit guidance on avoiding invented citations and on how to categorize each point (Expressly stated in user-provided materials; Expressly stated in verified online source; Supported but not expressly stated; Not found; Possible inference).
When to use it
Use this skill when evidential fidelity matters, such as reviewing documents, verifying claims, checking citations, producing chronologies, or performing legal research where gaps or inferences would be risky. It also triggers on specific phrases like 'source-locked' or 'verify this' to ensure a switch to source-locked mode.
What it can touch
The skill requires access to the user-provided materials and online sources actually accessed during the task. It uses the tool 'claude-code' as declared in the skill's metadata. All important points must be tied to precise references (document name + page/paragraph, URL + reference, etc.).
Caveats
All output must be anchored to explicit sources; no invented rules, numbers, or citations are allowed. If a fact cannot be verified from the provided materials or online sources actually checked, the agent must state that it was not found and indicate what would be needed to verify it. The system prohibits fabricating citations or asserting undisclosed details. The license is AGPL-3.0 as stated in the metadata.
# No Inference / Source-Locked Verification ## Purpose This skill exists because Claude's default behaviour is to be helpful — and being helpful often means filling gaps, making reasonable inferences, and providing complete-sounding answers. That default is dangerous when the user needs evidential fidelity. A plausible-sounding date that was never stated in the materials, a legal rule reconstructed from general knowledge rather than verified from the statute, a paragraph number that "looks right" — these are not helpful. They are liabilities. This skill forces Claude into a fundamentally different operating mode: **answer only from what you can see or have actually checked**. If it is not in the materials and Claude has not accessed an online source that states it, Claude does not state it as fact. Period. The skill is designed for legal, factual, research, evidence review, document review, citation-checking, chronology, and drafting tasks — any context where the user is relying on Claude's output as a faithful representation of what the sources actually say. --- ## Rule 1: Source-Locked Answers Only Claude must answer using only: - **materials provided by the user** (uploaded docu
- Purpose
- Rule 1: Source-Locked Answers Only
- Rule 2: When Claude Must Go Online
- Rule 3: No Unsupported Inferences
- Rule 4: Mandatory Evidential Anchoring
- Rule 5: Source Hierarchy
- Rule 6: Five Categories of Output
- Rule 7: Default Response to Missing Information
- Rule 8: No Invented Citations
- Rule 9: Quotations
- Rule 10: Dates and Deadlines
- Rule 11: Legal Rules and Propositions
- Rule 12: Appellate History and Case Status
- Rule 13: Conflict Handling
What does the source-locked-verification skill do?
No Inference / Source-Locked Verification. Forces Claude to answer ONLY from user-provided materials and/or online sources actually accessed — no inference, no assumptions, no gap-filling. Every factual, legal, numerical, or procedural claim must be anchored to a cited source. Use whenever Claude reviews documents, summarises evidence, checks accuracy, drafts submissions, creates timelines, extracts facts, checks citations, analyses rules, prepares legal arguments, compares documents, verifies claims, produces chronologies, works from uploaded materials, performs legal research, checks case st
How do I install it?
Run `npx skills add lawve-ai/awesome-legal-skills --skill source-locked-verification-larissa-meredith-flister --agent claude-code` — it drops the skill into your project so the agent can pick it up. Swap the --agent value for codex, cursor or copilot if you use one of those.
Where does this skill come from?
From lawve-ai/awesome-legal-skills, a repository with 618 stars. We read it straight from the repository tree rather than a submitted listing, so what you see here is what is actually published.
Is a popular skill a good skill?
Not necessarily. Stars measure attention, not adoption — a repository can trend for a week and be abandoned. That is why we show the weekly change from our own snapshots next to the total, instead of a single flattering number.