Skip to content

Caregiver Strategies

Status: build-out in progress. This page defines the bucket, its evidence base, taxonomy, and export contract. Individual strategy entries are promoted one at a time from trusted sources into runtime-ready micro-flows.

Strategies is GiveCare's third caregiver-support bucket, a sibling to benefits (programs you qualify for) and resources (orgs/places you can reach). It answers the caregiver's third question — what can I do? — with evidence-informed, non-clinical micro-support: problem-solving, micro self-care, communication scripts, help-seeking, and stress regulation.

It is not therapy and not "AI coaching." The defensible framing is micro-guided caregiver support: evidence-informed prompts, scripts, checklists, and one-next-step planning, with hard escalation boundaries.

Why this is its own bucket

The authoritative territory is real but the evidence is scattered across dementia caregiving, older-adult caregiving, hospice, disability, and self-management literature. A 2024 systematic review of caregiver self-care RCTs found the field still lacks a clean, practical self-care taxonomy5 — which is the gap GiveCare fills with a concrete micro-intervention taxonomy. The cleanest lane: micro self-care + problem-solving + communication + help-seeking + stress regulation.

The strongest blueprints are WHO iSupport1 (modular, self-guided, offline-friendly caregiver micro-lessons — closest to an SMS structure, justified as scalable public-health support2) and Problem-Solving Therapy / PISCES8 (the academic ancestor of "one next step"). The component lists in REACH II6, Powerful Tools for Caregivers7, FCA's Taking Care of YOU4, the NIA Caregiving Toolkit3, Savvy/Tele-Savvy9, the NYU Caregiver Intervention10, and VA HI-FIVES11 are the micro-primitives GiveCare borrows.

Taxonomy (the micro-intervention primitives)

Each strategy is a small, evidence-informed, do-it-yourself micro-flow. The category values below are the export contract. Authored runtime rows now exist for problem_definition, help_seeking, and one_next_step; the remaining non-escalation rows are the active build-out queue.

Category The micro-flow Source lineage
problem_definition "What is the actual problem today?" (name it in one sentence) PST/PISCES, REACH II
one_next_step goal → options → smallest safe action → follow-up reminder PST, Powerful Tools, FCA
stress_check identify warning signs, lower activation, take a small break FCA, NIA, REACH II
self_care sleep, food, movement, medical appointment, social contact NIA, WHO, Powerful Tools
help_seeking ask one person for one specific task FCA, Powerful Tools, NYUCI
communication script for family / clinician / benefits office / agency / employer Powerful Tools, NYUCI, HI-FIVES
reframing "what can you control?" / "what went slightly better?" (not therapy) REACH II, PST
preparedness what to ask at the next appointment, what documents to gather Savvy, HI-FIVES
resource_routing AAA, ADRC, Eldercare Locator, disease orgs, VA, benefits NIA, ACL-facing guidance
escalation safety boundary — crisis, abuse, unsafe transfer, medical change, mental-health risk safety layer (not a self-help tip)

The escalation category is a hard safety layer, not a self-help technique: it routes out of strategies to GiveCare's crisis path, never offers a "try this."

Export contract (strategies.json)

Like resources and benefits, the wiki owns this content and exports a bundled strategies.json that gc-sms reads in-memory through knowledge/catalog.ts and adapts through knowledge/toolAdapters.ts for agent/tools.ts — no Convex table or query round trip. One record per strategy:

{
  "id": "define-the-problem",
  "category": "problem_definition",
  "title": "Define the problem",
  "prompt": "What is the actual problem today? Name it in one sentence.",
  "steps": ["Name the problem in one sentence", "Sort urgent vs. not urgent", "Pick one goal"],
  "signals": ["Caregiver is overwhelmed by a tangled or vague problem"],
  "microSteps": [
    {
      "label": "Name one problem",
      "action": "Ask the caregiver to turn the tangled situation into one plain sentence.",
      "rationale": "Problem-solving starts by making one problem specific enough to act on.",
      "smsExample": "Try finishing this sentence: The hardest part today is __."
    }
  ],
  "example": "The hardest part today is getting Dad to Thursday appointments.",
  "avoid": ["Do not turn the strategy into a full care plan."],
  "followUp": "After the caregiver names the problem, offer one smallest safe action or one relevant resource lookup.",
  "safetyNotes": ["If the named problem includes immediate danger, use the safety path."],
  "whenToUse": "Caregiver is overwhelmed by a vague or tangled problem",
  "sourceIds": ["pst-pisces-hospice", "reach-ii"],
  "escalation": false
}

escalation: true marks the safety-boundary entry. steps and whenToUse are optional. sourceIds must reference real docs/sources/<id>.md entries.

Build-out status

The rails are built: the strategy content type + schema (scripts/schema.ts), the extract→publish emit to data/exports/strategies.json (scripts/publish.ts), and the validating gc-sms sync/consumer (scripts/sync-wiki.ts, knowledge/catalog.ts, plus the knowledge/toolAdapters.ts strategy lookup). Current authored non-escalation strategies: define-the-problem, ask-for-one-specific-help, and one-next-step.

What remains is content build-out for the other taxonomy rows, with each page carrying runtime signals, micro-steps, examples, avoid rules, follow-up behavior, safety notes, and cited source IDs.

Authoring a strategy

Add a page at docs/strategies/<slug>.md, copying define-the-problem.md:

---
title: <short, plain title>
description: <one line>
type: strategy
strategy:
  category: <one taxonomy category above>
  prompt: "<the SMS-sized micro-prompt / one question>"
  steps:               # optional ordered micro-steps
    - <step>
  signals:             # concrete runtime triggers
    - <when this strategy fits>
  microSteps:          # unpacked runtime guidance
    - label: <short label>
      action: <what Mira should do>
      rationale: <why this step exists>
      smsExample: <optional SMS-sized wording>
  example: <sample caregiver-facing wording or ask>
  avoid:
    - <unsafe or unhelpful pattern to avoid>
  followUp: <what to do after the caregiver responds>
  safetyNotes:
    - <when to leave strategy lane and escalate or use another tool>
  whenToUse: <short trigger>
  escalation: false    # true ONLY for a crisis-routing boundary entry
sources:               # cite >=1 real docs/sources/<id>.md
  - id: <source-id>
    cited_for: <what it backs>
tags: [strategy]
updated: <YYYY-MM-DD>
updated_by: <you>
---

# <title>

<plain, strength-framed body. Not therapy, not medical advice.>

Then emit + verify:

hound plan --driver research/hound-driver.json --operation corpus.project --json '{}' \
  --as-of YYYY-MM-DD --output .hound/project-plan.json
hound execute --driver research/hound-driver.json \
  --plan .hound/project-plan.json  # validates and writes owner exports
(cd ../gc-sms && pnpm sync:wiki)      # validates and copies runtime bundles
pnpm lint                             # must stay 0 errors

Guardrails (non-negotiable):

  • Non-clinical only — no diagnoses, dosages, or medical procedures.
  • escalation is a safety boundary, not a tip. gc-sms filters escalation entries out of suggestions; crisis routing (988/911) is the harness policy path. Don't author escalation entries as self-help; if you do, set escalation: true so they're excluded from lookup.
  • Cite real sources in docs/sources/; never invent evidence.
  • One concrete micro-flow per page; keep prompts SMS-sized.

  1. WHO. "iSupport for Dementia." Full entry → ↩

  2. Pot AM, et al. "iSupport: a WHO global online intervention for informal caregivers of people with dementia." Full entry → ↩

  3. NIA. "Caregiving Toolkit." Full entry → ↩

  4. Family Caregiver Alliance. "Taking Care of YOU: Self-Care for Family Caregivers." Full entry → ↩

  5. BMC Geriatrics. "RCTs on promoting self-care behaviors among informal caregivers of older patients: a systematic review and meta-analysis." Full entry → ↩

  6. Belle SH, et al. "REACH II: Resources for Enhancing Alzheimer's Caregiver Health." Full entry → ↩

  7. "Powerful Tools for Caregivers." Full entry → ↩

  8. Demiris G, et al. "A Problem-Solving Intervention for Hospice Family Caregivers (PISCES): A Randomized Clinical Trial." Full entry → ↩

  9. Kally Z, et al. "Savvy Caregiver / Tele-Savvy." Full entry → ↩

  10. Mittelman MS, et al. "NYU Caregiver Intervention." Full entry → ↩

  11. Van Houtven CH, et al. "HI-FIVES: Family Caregiver Skills Training to Improve Experiences of Care." Full entry → ↩