CtrlK
BlogDocsLog inGet started
Tessl Logo
SkillPublic

phi-prompt-guard

Runtime, prompt-time behavioral guardrail that helps reduce PHI exposure in LLM-assisted workflows by detecting PHI-bearing prompts, avoiding unsafe tool actions that would pull more PHI in, and redirecting users toward de-identified or synthetic inputs. Use when the user is about to paste, query, or read clinical/patient data, or when an action (DB query, file read, tool output) may pull PHI into the conversation. Honors a [PHI-OK] attestation for synthetic / test data.

Quality

84%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide
SecuritybySnyk

Passed

No findings from the security scan

SKILL.md
Quality
Evals
Security

Quality

Content

75%Weight 40%Scale 1-5

Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.

The body is an unusually actionable behavioral policy — concrete commands, exact example replies, and decision rules with named exceptions — but it says the same things multiple times across its sections and could shed 30-40% of its length without losing any decision-relevant content. Consolidating the repeated [PHI-OK] and generate-don't-execute material into single authoritative sections would tighten it substantially.

Suggestions

Consolidate the generate-don't-execute guidance, which currently appears in the Workflow (step 5), Key Features, the Actions table, and the Operating rules, into one authoritative statement with the Actions table as its sole supporting detail.

Merge the three [PHI-OK] discussions (Workflow step 3, Instructions item 3, and 'The [PHI-OK] Attestation' section) into the dedicated attestation section and reference it from the workflow, removing the duplicated default-behavior and override rules.

Drop the 'Dependencies: None' section and the repeated 'Pure in-context behavioral skill — no scripts, no external services' framing (already stated in the intro), and consider moving the 18-identifier taxonomy and override-signal checklist to a references/ file to shorten the main skill body.

DimensionReasoningScore

Conciseness

Mostly efficient — the operating rules, examples, and decision heuristics earn their tokens — but there is real repetition: "generate, don't execute" is stated in the Workflow, Key Features, Actions table, and Operating rules; the [PHI-OK] attestation is explained in three separate sections; and "Pure in-context behavioral skill — no scripts, no external services" plus an empty "Dependencies: None" section pad the file. The 18-identifier taxonomy is also largely knowledge Claude already has, though it adds operational specifics (ZIP3, 90+ rule).

3 / 5

Actionability

Fully actionable for an instruction-only skill: exact database clients and dump tools named, safe inspection commands given verbatim (`head -1`, `\dt`, `SHOW TABLES`, `wc -l`), concrete model replies to copy in each scenario, precise decision rules ("schema is safe, data is not"; "unknown environment ⇒ assume production"), and explicit override signals for the [PHI-OK] case — no vague direction anywhere.

5 / 5

Workflow Clarity

The 6-step per-prompt workflow is clearly sequenced with genuine checkpoints (assess before every prompt, override the attestation only on clustered real-chart signals, ask the user to confirm when in doubt), and the no-[PHI-OK] response procedure includes feedback loops (redact and re-submit). Held at 4 rather than 5 because the decision logic is fragmented across four sections — Workflow, Instructions, The [PHI-OK] Attestation, and When PHI Is Detected — so the reader must reassemble the branch tree.

4 / 5

Progressive Disclosure

A single self-contained file with well-organized, clearly headed sections and no buried or nested references, which is appropriate for a documentation-only behavioral skill with zero dependencies. Minor gap: at ~165 lines, reference-like material (the 18-identifier taxonomy and the [PHI-OK] override heuristics) is inlined in SKILL.md where the pattern favors splitting into a references/ file, keeping the overview leaner.

4 / 5

Total

16

/

20

Passed

Description

87%Weight 40%Scale 1-5

Based on the skill's description, can an agent find and select it at the right time? Clear, specific descriptions lead to better discovery.

A strong description: third-person voice, explicit what-and-when with concrete trigger phrases, and a clearly distinct niche anchored by the memorable [PHI-OK] attestation. The only gap is a handful of natural synonyms (HIPAA, EHR, MRN) that live in the body rather than the description itself.

DimensionReasoningScore

Specificity

The description lists several concrete actions — "detecting PHI-bearing prompts", "avoiding unsafe tool actions that would pull more PHI in", "redirecting users toward de-identified or synthetic inputs", plus the "[PHI-OK] attestation" mechanism — but the actions are stated at a behavioral-policy altitude ("avoiding", "redirecting") rather than the fully concrete operation level of the 5 anchor ("extract text, fill forms, merge documents, convert pages"), leaving minor coverage gaps.

4 / 5

Completeness

Both halves are explicit: the "what" is a "Runtime, prompt-time behavioral guardrail that... detect[s], avoid[s], redirect[s]", and the "when" is a concrete trigger clause — "Use when the user is about to paste, query, or read clinical/patient data, or when an action (DB query, file read, tool output) may pull PHI into the conversation" — matching the 5 anchor's what-plus-when-with-triggers pattern.

5 / 5

Trigger Term Quality

Good keyword coverage with natural user phrasing — "paste, query, or read clinical/patient data", "DB query, file read, tool output" — but misses common synonyms users would plausibly say such as HIPAA, EHR, medical records, or MRN, which only appear in the body, not the description.

4 / 5

Distinctiveness Conflict Risk

The PHI/HIPAA clinical-data guardrail niche with its distinctive "[PHI-OK]" token has minimal overlap with any other skill category; the trigger conditions (clinical/patient data, PHI-pulling actions) are highly specific and would not fire for unrelated skills.

5 / 5

Total

18

/

20

Passed

Validation

93%

Checks the skill against the spec for correct structure and formatting. All validation checks must pass before discovery and implementation can be scored.

Validation — 15 / 16 Passed

Validation for skill structure

CriteriaDescriptionResult

frontmatter_unknown_keys

Unknown frontmatter key(s) found; consider removing or moving to metadata

Warning

Total

15

/

16

Passed

Repository

aipoch/medical-research-skills

Path

scientific-skills/Other/phi-prompt-guard/SKILL.md

Reviewed

Table of Contents

Is this your skill?

If you maintain this skill, you can claim it as your own. Once claimed, you can manage eval scenarios, bundle related skills, attach documentation or rules, and ensure cross-agent compatibility.