| name | mental-health-care-plan |
| description | Draft, audit, and manage patient care plans — synthesize assessment results into structured treatment plans, and evaluate existing plans for clinical gaps, safety concerns, and grounding issues. |
| version | 0.2.0 |
| author | hermes-mental-health |
| license | MIT |
| metadata | {"hermes":{"tags":["mental-health","care-plan","treatment","clinical","audit"],"category":"mental-health","related_skills":["mental-health-core","mental-health-assessment-review","mental-health-patient-summary"]}} |
Mental Health Care Plan
When to use
Use when the practitioner needs to:
- Draft a care plan based on assessment results
- Audit an existing care plan — evaluate against assessment results, check goal grounding, flag phantom baselines, identify missing interventions, and score quality
- Define treatment goals and interventions
- Schedule follow-up assessments
- Document clinical rationale and expected outcomes
- Document clinical rationale and expected outcomes
Care plan structure
Draft workflow
Sections (in order):
- Treatment Goals — numbered list, each goal measurable and time-bound with specific score targets
- Recommended Interventions — bullet list, each with modality, frequency, and timeline
- Follow-up Plan — bullet list of measures, frequency, and key review milestones
- Discharge Criteria — bullet list of concrete, verifiable exit conditions
- Risk Mitigation — bullet list covering suicide screening, safety planning, substance use, non-response flags, and missed session protocols
Audit workflow
When auditing an existing care plan (not drafting from scratch):
- Load all assessment results from
data/patients/<id>/results/ — every goal must be grounded in a real result file
- Verify measure versions — check that cited measures match what was actually administered (e.g., adult vs child version)
- Check for phantom baselines — goals referencing PHQ-9, GAD-7, or other scores that have no result file on disk
- Validate time-sensitive data — medication doses, appointment dates, assessment due dates against current date
- Cross-reference clinical background — ensure interventions match documented history (prior med trials, family history, substance use)
- Score the plan — return: overall quality score (0-100), strengths, clinical gaps, measurable-goal issues, missing interventions, safety concerns, and recommended revisions
Audit output sections:
- Overall quality score (numeric + rating)
- Strengths
- Clinical gaps (each rated Critical/Moderate/Minor)
- Measurable-goal issues (tabulated: goal, issue, severity)
- Missing evidence-based interventions
- Safety concerns
- Recommended revisions (grouped: immediate, short-term, medium-term)
Formatting constraints
ONLY these markdown elements are permitted:
-
and ### headings
- Bullet lists (-)
- Numbered lists (1.)
- bold
- --- horizontal rules
Never use: HTML, tables, code blocks, blockquotes, nested blocks, or special Unicode beyond basic ASCII.
Data grounding
Every care plan must be grounded in the patient's actual assessment results. Read the latest results from data/patients/<id>/results/ and the clinical-background.md before drafting. Goals must reference specific baseline scores from real assessment data. Never generate a plan from only the outline or prior plan summary.
Common audit findings (pitfalls to check)
- Phantom baselines — goals reference PHQ-9/GAD-7 scores that were never collected. Verify every score target has a corresponding result file.
- Wrong measure version — plan cites "Adult version" but only a Child 11-17 result exists. Check the result's
assessmentSlug against the plan's claims.
- Safety gate deferred — plan gates suicide screening behind PHQ-9 item 9 ≥2 but no PHQ-9 has been administered. For patients with passive ideation + family history, baseline C-SSRS should be done at session 1 regardless.
- Missing baselines — AUDIT-C, DAST-10, C-SSRS, or somatic measures (PHQ-15/SSS-8) are scheduled but never administered.
- Time-sensitive staleness — medication doses, appointment dates, or week-based milestones that have passed without update.
Care Plan Audit
When the practitioner requests an audit of existing care plans, do NOT draft a new plan from scratch. Evaluate the existing care plan against the patient's actual assessment results, clinical background, and safety profile.
Audit procedure
- Load all assessment results from
data/patients/<id>/results/ — read every JSON file
- Load the existing care plan from
data/patients/<id>/care-plan.md
- Load clinical background from
data/patients/<id>/clinical-background.md
- Cross-reference every goal against actual result files — flag any goal that references a score from a measure that was never administered (phantom baselines)
- Verify measure versions — confirm the measure version in the plan matches the actual result file (e.g., adult vs child version of Level 1)
- Check safety gaps — suicide risk stratification, safety plan timing, emergency contact documentation
- Check missing assessments — identify assessments the plan references but were never collected (PHQ-9, GAD-7, AUDIT-C, DAST-10, C-SSRS)
- Evaluate goal measurability — each goal must have a concrete score target, timeline, and tracking mechanism
- Check intervention evidence basis — interventions should match gold-standard protocols for the condition (e.g., CBT-I for insomnia, not just sleep hygiene)
- Score the plan — assign an overall quality score (0-100) and rating (Excellent/Good/Needs Revision/Unreliable)
Audit output format
# Care Plan Audit — <patient-id>
**Audit date:** <date>
**Plan under review:** care-plan.md (generated <date>)
**Assessment results on file:** <N> (<list slugs>)
---
## Overall Quality Score: <N>/100
**Rating:** <Excellent | Good | Needs Major Revision | Unreliable>
---
## Strengths
[bullet list of what the plan does well]
---
## Clinical Gaps (Critical)
### GAP 1 — <title> (Critical)
[description, impact, recommended revision]
---
## Measurable-Goal Issues
[table: Goal | Issue | Severity]
---
## Missing Evidence-Based Interventions
[bullet list]
---
## Safety Concerns
[numbered list]
---
## Recommended Revisions (Summary)
### Immediate
### Short-term
### Medium-term
---
## Bottom Line
[one-sentence verdict]
Common audit findings (pitfalls to check)
- Wrong measure version — Child/adolescent measure administered to an adult (or vice versa); plan references "Adult version" but result file is child version
- Phantom baselines — Plan sets PHQ-9/GAD-7 targets but no PHQ-9/GAD-7 result file exists on disk
- Safety plan deferred — Safety plan scheduled at session 2+ for a patient with documented passive death ideation and family history of suicide
- Suicide risk gated behind screening — C-SSRS deferred behind PHQ-9 item 9 trigger when baseline suicide risk is already elevated
- Somatic discrepancy unvalidated — Self-report vs clinician-rating gap noted but no validated somatic measure (PHQ-15, SSS-8) scheduled
- Substance use baselines missing — AUDIT-C/DAST-10 mentioned in risk mitigation but never administered
- Sleep hygiene only — Sleep hygiene recommended for depression-linked insomnia when CBT-I (stimulus control, sleep restriction) is gold standard
- Couples/family component underspecified — Mentions spouse involvement but no session allocation, modality, or outcome measure
Data sources
All care plan data comes from local sources:
- Profile data from
data/patients/<id>/profile.json
- Assessment results from
data/patients/<id>/results/
- Clinical background from
data/patients/<id>/clinical-background.md
- Clinical summary from
data/patients/<id>/clinical-summary.md
Storage
Care plans are stored in data/patients/<id>/care-plan.md.
Versioned backups are stored in data/patients/<id>/version/care-plan-<ts>.md before each overwrite.
Never store care plans outside the patient's scoped directory.
Agent prompt mapping
The "Result — Audit Care Plan" prompt in lib/prompts.ts maps to agent care-plan, which loads skills mental-health-core,mental-health-care-plan (defined in AGENT_SKILLS in app/(dashboard)/_components/agent-modal.tsx). The prompt instructs the agent to audit existing plans, not draft from scratch — see the data grounding and audit sections above.
References
references/care-plan-example-josoroma.md — concrete example with baseline scores, targets, and grounded interventions
references/care-plan-audit-pitfalls.md — detailed audit failure modes with real examples from josoroma-mqn4h6m8 audit (June 2026)