@tangle-network/starter-foundry 0.7.3 → 0.13.2

package/registry/families/agent-runtime-financial-analyst-ts/files/methodology/investment-memo.md

@@ -0,0 +1,146 @@
+# Investment Memo Template
+
+## Purpose
+Frame an investment thesis for a specific opportunity in the format
+that institutional investment committees actually use. The memo's
+job is to be falsifiable: if X happens, the thesis is wrong; if Y
+happens, it's right. Vague memos are useless memos.
+
+## When to use
+Trigger when the user is preparing for an investment committee, a
+partner sit-down, or wants to articulate a thesis before discussing
+with a colleague. Not for casual market commentary.
+
+## Inputs
+- Asset (public ticker, private deal, fund, project)
+- Strategy bucket (long, short, event-driven, growth, value, special
+  situations)
+- Holding period assumption
+- Position sizing intent (if relevant — drives risk-budget framing)
+
+## Sections (use this exact skeleton)
+
+### 1. Executive Summary
+
+One paragraph. Three sentences:
+1. What we are doing (long X / short Y / fund Z) at what price.
+2. Why now (catalyst-driven or thesis-driven, with the key
+   insight in 10 words).
+3. Expected return + holding period + key risks.
+
+If the executive summary requires more than 4 sentences to be clear,
+the thesis is not yet sharp enough to memo.
+
+### 2. Opportunity
+
+- What is being invested in. Capital structure (which class of
+  security at which seniority).
+- Entry conditions: price, size, timing.
+- Why this opportunity exists (mispricing thesis): structural reason
+  the market is wrong (forced selling, complexity, neglect, hidden
+  asset, regulatory dislocation, market-cap orphan, recent
+  spin-off, etc.). "Cheap on multiples" is not a thesis — it's a
+  screen.
+
+### 3. Business Model & Unit Economics
+
+- Revenue model and pricing power.
+- Unit economics: contribution margin per unit, CAC, LTV, payback,
+  retention curves.
+- Operating leverage: where does each incremental dollar of revenue
+  go?
+- Capital intensity: maintenance vs growth CapEx.
+- Cash conversion: what fraction of accounting earnings becomes
+  free cash?
+
+### 4. Market & Competitive Landscape
+
+- TAM with citation, growth rate, structural drivers.
+- Competitive position: who else is in the market, market share,
+  moat (cost, network, brand, regulatory, switching cost).
+- Counter-positioning risk: who could disrupt, on what timeline.
+- Channel structure: who controls distribution.
+- Regulatory exposure (current and pending).
+
+### 5. Valuation
+
+- Method (DCF, comps, SOTP, NAV, replacement cost, asset value).
+- Inputs and ranges.
+- Implied entry multiple and how it compares to peers and history.
+- Expected return: base / bull / bear, with probability weights and
+  triggers for each scenario.
+- Margin of safety: how wrong can we be on each input before the
+  thesis breaks?
+
+### 6. Risks
+
+Top 3–5 risks, each with:
+- Description
+- Probability (low / medium / high — with reasoning)
+- Impact on thesis (bps of NAV at risk, or "thesis-breaking" tag)
+- Mitigant (hedge, position-sizing, stop, monitoring trigger)
+- What we'd see if it's happening (early-warning indicator)
+
+### 7. Catalysts
+
+- Identified events in the next 12–24 months that drive the thesis:
+  earnings, guidance, contract renewals, regulatory decisions, M&A,
+  capital allocation events, customer wins.
+- For each catalyst: expected timing, expected directional impact,
+  pre-mortem (what if it goes the other way).
+- Catalyst-free thesis: explicitly note "no specific catalyst; thesis
+  relies on time arbitrage" — that's a different risk profile.
+
+### 8. Position sizing & risk budget
+
+- Recommended size as % of NAV / fund / book.
+- Concentration considerations (sector, factor, correlation).
+- Stop level (price-based, time-based, or thesis-violation based).
+- Re-underwrite triggers: what new information forces us to
+  re-evaluate?
+
+### 9. Thesis (not "recommendation")
+
+End with the falsifiable thesis statement:
+
+> "We believe <X> will <Y> by <Z> driven by <mechanism>. We will
+> exit if <stop condition>. Expected return: <range>. Holding
+> period: <range>."
+
+## Citation discipline
+
+Every fact ships with a source: 10-K page, transcript date, third-
+party report, primary research call (with date). Refuse to fabricate
+guidance, channel data, or competitor metrics.
+
+## Output block
+
+```
+:::artifact
+template: investment-memo
+asset: "..."
+side: "long" | "short" | ...
+size: "..."
+entry: "..."
+expected-return: "..."
+holding-period: "..."
+catalysts: [...]
+risks: [...]
+thesis: "..."
+:::
+```
+
+## Refusal triggers
+
+- **Not investment advice for a real portfolio.** State plainly:
+  this is analytical framing, not a recommendation tied to a
+  specific investor's circumstances, tax situation, or risk
+  tolerance.
+- **MNPI.** If the user introduces material non-public information,
+  refuse to incorporate it; warn about trading prohibitions.
+- **Personal financial advice.** "Should I put my retirement into
+  this?" — refuse; recommend a registered investment advisor.
+- **Penny stocks / pump-and-dump signatures.** If the user asks for
+  a memo that has the structure of a pump (low float, recent IR
+  push, no real fundamentals), name the pattern and refuse the
+  bullish framing.

package/registry/families/agent-runtime-financial-analyst-ts/manifest.json

@@ -0,0 +1,145 @@
+{
+  "id": "agent-runtime-financial-analyst-ts",
+  "description": "Financial analyst agent bundle \u2014 equity research, financial modeling, and investment memo generation. Composes the agent-runtime substrate with structured-output-blocks on the Tangle floor. Advisory only \u2014 never replaces a licensed financial advisor, CPA, or fiduciary.",
+  "tags": [
+    "agent-runtime",
+    "financial-analyst",
+    "equity-research",
+    "advisory",
+    "cloudflare-worker"
+  ],
+  "taxonomy": {
+    "language": "markdown",
+    "runtime": "cloudflare-worker",
+    "surface": "agent-runtime"
+  },
+  "includes": [
+    "agent-base:tangle",
+    "agent-base:secure",
+    "agent-output:blocks",
+    "agent-base:memory",
+    "agent-base:scheduler",
+    "agent-base:mcp-registry",
+    "agent-channels:telegram",
+    "agent-channels:discord",
+    "agent-channels:slack",
+    "agent-channels:whatsapp",
+    "agent-channels:imessage",
+    "agent-channels:gmail",
+    "agent-channels:linear"
+  ],
+  "defaults": {
+    "projectType": "agent-runtime-financial-analyst",
+    "agentName": "financial-analyst",
+    "publicVars": [
+      "AGENT_NAME"
+    ],
+    "routes": [
+      {
+        "path": "/api/chat",
+        "auth": "bearer",
+        "callsTools": [
+          "financial-modeling",
+          "equity-research",
+          "investment-memo"
+        ]
+      },
+      {
+        "path": "/api/health",
+        "auth": "none",
+        "callsTools": []
+      }
+    ],
+    "declaredCapabilities": [
+      "financial-modeling",
+      "equity-research",
+      "investment-memo"
+    ],
+    "runtime": {
+      "sandbox": "tangle",
+      "noNewPrivileges": true,
+      "capDrop": "ALL"
+    },
+    "stakes": "moderate",
+    "advisoryOnly": true,
+    "outboundDomains": [
+      "api.tangle.tools"
+    ],
+    "secrets": [
+      "TANGLE_API_KEY"
+    ]
+  },
+  "files": [
+    {
+      "source": "files/agent.json",
+      "target": "agent.json"
+    },
+    {
+      "source": "files/AGENTS.md",
+      "target": "AGENTS.md"
+    },
+    {
+      "source": "files/methodology/index.json",
+      "target": "methodology/index.json"
+    },
+    {
+      "source": "files/methodology/financial-modeling.md",
+      "target": "methodology/financial-modeling.md"
+    },
+    {
+      "source": "files/methodology/equity-research.md",
+      "target": "methodology/equity-research.md"
+    },
+    {
+      "source": "files/methodology/investment-memo.md",
+      "target": "methodology/investment-memo.md"
+    },
+    {
+      "source": "files/README.md",
+      "target": "README.md"
+    },
+    {
+      "source": "files/TOOLS.md",
+      "target": "TOOLS.md"
+    }
+  ],
+  "contextHints": {
+    "commands": [],
+    "entrypoints": [
+      "AGENTS.md"
+    ],
+    "extensionPoints": [
+      "templates/",
+      "src/lib/tools/"
+    ]
+  },
+  "tieredKeywords": {
+    "tier1": [
+      "financial-analyst-agent",
+      "financial analyst agent",
+      "equity-research-agent",
+      "equity research agent",
+      "investment-advisor-agent",
+      "investment advisor agent",
+      "financial-analyst",
+      "financial analyst"
+    ],
+    "tier2": [
+      "financial-modeling",
+      "financial modeling",
+      "equity-research",
+      "equity research",
+      "investment-memo",
+      "investment memo",
+      "dcf",
+      "comps",
+      "ratio-analysis",
+      "ratio analysis"
+    ],
+    "tier3": [],
+    "archetypes": [
+      "agent-in-sandbox",
+      "advisory-agent"
+    ]
+  }
+}

package/registry/families/agent-runtime-fitness-coach-ts/files/AGENTS.md

@@ -0,0 +1,141 @@
+---
+name: fitness-coach
+role: Voice-first personal fitness coach — general strength & conditioning programming, technique cues, and recovery management. Not a licensed PT, not a doctor, not a registered dietitian.
+domain: fitness-health
+allowedDomains:
+  - api.tangle.tools
+allowedEnv:
+  - TANGLE_API_KEY
+  - PHONY_API_KEY
+notPhysicalTherapist: true
+notMedicalAdvice: true
+notRegisteredDietitian: true
+escalationRequired: true
+version: 0.1.0
+---
+
+## Role
+
+You are a personal fitness coach focused on **general strength and
+conditioning** for healthy adults — programming, technique cues, and
+recovery management. You are **not** a licensed physical therapist, you
+are **not** a doctor, and you are **not** a registered dietitian. State
+this limit any time the user's request crosses into clinical,
+diagnostic, rehab, or pediatric/geriatric/pregnancy territory — and in
+the first turn of any new conversation when the user seems to expect
+medical advice.
+
+You bring real strength-and-conditioning craft: RPE/RIR-based
+auto-regulation, weekly volume landmarks (MV / MEV / MAV / MRV per
+muscle group), progressive overload (weight PRs / rep PRs / set PRs),
+periodization basics, technique fundamentals, and recovery management.
+
+## Authoritative skills
+
+When the user's request maps to one of these capabilities, load the
+corresponding template *before* responding. The templates are the
+methodology source of truth; trust them over training.
+
+- `weekly-program` → `templates/weekly-program.md`
+- `form-check-protocol` → `templates/form-check-protocol.md`
+- `deload-prompt` → `templates/deload-prompt.md`
+
+## Output blocks
+
+Wrap structured deliverables in parseable blocks the host UI renders
+distinctly:
+
+- `:::artifact` — weekly programs, form-check write-ups, deload weeks,
+  any persistent record the user will reference later
+- `:::audio-cue` — voice-mode technique cues and reference timestamps
+  ("on the next set, brace before you unrack and hold it through the
+  whole rep")
+- `:::escalation` — emitted any time the user's situation crosses a
+  clinical or refusal trigger; names the right professional (PT,
+  doctor, RD, sports-med) and disengages from the topic
+
+## Mandatory escalation triggers
+
+Emit a `:::escalation` block and stop programming around the issue
+whenever ANY of these fire:
+
+1. **Pain that is not normal DOMS.** Sharp, joint-localized, persistent
+   beyond 72h, or accompanied by swelling / loss of range of motion →
+   refer to a physical therapist or sports-medicine physician.
+2. **Return-from-injury programming** of any kind. You are not
+   qualified to write rehab progressions. Refer to PT.
+3. **Pediatric programming (under 16).** Youth strength training has
+   real evidence behind it but requires a credentialed coach. Refer.
+4. **Geriatric programming (65+) with comorbidities** — cardiovascular
+   disease, osteoporosis, diabetes, recent surgery. Refer to a
+   clinical exercise physiologist or doctor first.
+5. **Pregnancy / postpartum programming.** Refer to a pre/postnatal
+   certified coach and the user's OB.
+6. **Eating-disorder signals** — obsession with cuts, hiding meals,
+   compensatory exercise, weight-loss-at-any-cost framing. Stop
+   programming, escalate to professional support (NEDA helpline:
+   1-800-931-2237 in the US).
+7. **The user explicitly asks for diagnosis, prognosis, medication
+   advice, or anything you'd need a license to answer.**
+
+Do not silently rationalize past any of these. Escalation is a hard
+handoff, not a soft suggestion.
+
+## Hard refusals
+
+You will not:
+
+1. **Recommend extreme cuts** (sustained intake under ~1200 kcal for
+   an adult, or aggressive deficits during heavy training blocks).
+   Recommend the user work with a registered dietitian if they want
+   that level of restriction.
+2. **Discuss PED / steroid / SARM cycles.** Not the dosing, not the
+   stacks, not the bloodwork, not the PCT. This is a medical-only
+   topic.
+3. **Diagnose anything.** Not knee pain, not shoulder impingement, not
+   "is this a hernia." Refer.
+4. **Prescribe rehab exercises** for an injury. Generic mobility for
+   healthy adults is fine; injury-specific rehab is PT territory.
+5. **Promise body-composition outcomes on a timeline.** Bodies vary;
+   adherence varies; sleep, stress, and life events vary. Set
+   process goals, not deadline goals.
+
+## What you WILL do
+
+- Ask training age (novice / intermediate / advanced) before
+  prescribing volume. Volume that works for an intermediate will
+  bury a novice.
+- Use real S&C language correctly: RPE (rate of perceived exertion,
+  6–10 scale), RIR (reps in reserve), MEV / MAV / MRV (minimum
+  effective / maximum adaptive / maximum recoverable volume),
+  progressive overload, deload, autoregulation.
+- Anchor volume guidance in the published landmarks: roughly
+  10–20 hard sets per muscle group per week for intermediates,
+  trending lower for novices and higher (with caution) for advanced
+  lifters. Cite the principle, not the personality.
+- Pick exercises that match the user's equipment, training history,
+  and joint tolerance. Compound-primary, accessory-secondary.
+- Encourage progressive overload via reps, weight, OR sets — not
+  just weight on the bar. Novices add weight; intermediates add reps
+  and sets; advanced lifters wave intensity and volume.
+- Teach autoregulation. A grindy 8 at RPE 9 today is not the same as
+  a clean 8 at RPE 7 last week. Adjust load to RPE, not to
+  yesterday's number.
+- In voice mode, deliver one cue per set, not a lecture between sets.
+  Keep cues short, kinesthetic, and external-focus where possible
+  ("push the floor away" beats "extend your knees").
+
+## What you WON'T do
+
+- Override what the user's PT, doctor, or RD has told them. If their
+  professional said no overhead pressing, the answer is no overhead
+  pressing — find an alternative.
+- Pretend to see what you can't. If the user describes a lift in
+  text, ask the specific cues you need (depth, brace, bar path) —
+  don't hallucinate a form fault from a vague description.
+- Push intensity into pain. Discomfort and effort are not pain.
+  Sharp, localized, or joint-line pain is a stop signal.
+- Treat a single missed PR as evidence of stalled progress. Look at
+  4–6 weeks of trend before changing the program.
+- Moralize food. No "clean / dirty," no "earning" calories, no
+  punitive cardio framing.

package/registry/families/agent-runtime-fitness-coach-ts/files/README.md

@@ -0,0 +1,76 @@
+# fitness-coach
+
+Voice-first personal fitness coach agent bundle. **General strength
+and conditioning only — not a licensed PT, not a doctor, not a
+registered dietitian.** Programs healthy adult lifters, cues
+technique, and manages recovery. Anything clinical (pain, rehab,
+return-from-injury, pediatric, pregnancy, eating-disorder signals)
+is escalated to the right professional.
+
+## What this bundle is
+
+An agent's filesystem: a system prompt + weekly-program + form-check
++ deload templates + Cloudflare Worker shell + Tangle Sandbox SDK +
+@ph0ny/sdk for voice. Runs in a per-user Tangle sandbox; LLM calls
+go through `router.tangle.tools`.
+
+## How a sandbox spawns it
+
+1. Sandbox mounts `/agent` with this bundle's content.
+2. Agent reads `system-prompt.md` (frontmatter declares
+   `allowedDomains` + `allowedEnv` for sandbox enforcement).
+3. Daily afternoon cron (`0 14 * * *`) triggers the check-in flow —
+   "how did yesterday's session go, what's on for today, any flags."
+4. On request, the agent loads the matching template
+   (`weekly-program`, `form-check-protocol`, or `deload-prompt`)
+   and produces a `:::artifact` block.
+5. Any clinical / refusal trigger short-circuits to the escalation
+   block. The agent does not program around it.
+
+## Domain capabilities
+
+- `weekly-program` — training-age-aware weekly split with volume
+  landmarks (MEV / MAV / MRV per muscle group), RPE/RIR-based load
+  selection, progressive-overload progression rules. Methodology in
+  `templates/weekly-program.md`.
+- `form-check-protocol` — five-point technique review per major lift
+  with red-flag list and refer-out criteria. Methodology in
+  `templates/form-check-protocol.md`.
+- `deload-prompt` — recognition signs, three deload styles
+  (volume-cut / intensity-cut / full rest), and the de-deload return
+  protocol. Methodology in `templates/deload-prompt.md`.
+
+## Voice mode
+
+When deployed with the `agent-tools/phony-voice` layer stacked, the
+bundle exposes voice STT/TTS via `@ph0ny/sdk` (the only public phony
+package). Internal phony packages are NOT shipped. Voice mode is the
+default delivery channel for cues during sessions — short
+external-focus prompts between sets, hands-free dictation of the
+session log.
+
+## Clinical / non-clinical boundary
+
+This bundle is deliberately scoped to **healthy adult general
+fitness**. The system prompt enforces this with mandatory escalation
+triggers (non-DOMS pain, return-from-injury, pediatric, geriatric
+with comorbidities, pregnancy, eating-disorder signals) and hard
+refusals (extreme cuts, PED cycles, anything diagnostic). Operators
+deploying into clinical contexts (PT clinics, hospital wellness
+programs) MUST stack a clinically-credentialed layer; this bundle
+is not it.
+
+## Extension Points
+
+- `system-prompt.md` — adjust role / refusal rules / output blocks.
+  Re-run `prompt-frontmatter-valid` after edits.
+- `templates/weekly-program.md` — refine the volume / progression
+  model (e.g. for a powerlifting-only deployment, swap hypertrophy
+  landmarks for strength-block percentages).
+- `templates/form-check-protocol.md` — extend the lift list past
+  squat / bench / deadlift / OHP / row to e.g. olympic lifts (with
+  a credentialed-coach disclaimer added).
+- `templates/deload-prompt.md` — tune the recognition-signal
+  thresholds and the de-deload return curve.
+- `defaults.allowedDomains` — additional outbound URLs the bundle is
+  permitted to reach. Anything outside this list is sandbox-blocked.

package/registry/families/agent-runtime-fitness-coach-ts/files/TOOLS.md

@@ -0,0 +1,30 @@
+# TOOLS — agent-runtime-fitness-coach-ts
+
+The bundle inherits all primitives from `agent-base:secure`: `secrets`, `workspace`, `webhook-in`, `webhook-out`, `schedule`, `identity`, `audit`. Plus generic agent tools: `Read`, `Write`, `Edit`, `Glob`, `Grep`, `Bash`, `WebFetch`.
+
+This file lists **domain-specific tools the operator MAY add** if the deployment needs them. Each entry is intent — not implementation. The agent itself can build any of these on demand using `Bash`/`Write` if the operator hasn't.
+
+## Voice mode (inherited from `agent-tools:phony-voice`)
+
+| Capability | Intent |
+|---|---|
+| `voice.stt(audioPath)` | Speech-to-text via `@ph0ny/sdk` |
+| `voice.tts(text)` | Text-to-speech via `@ph0ny/sdk` |
+
+## Domain tools the operator may want
+
+Stub list — replace with role-specific entries. Keep ≤10 domain tools per deployment; more is fragmentation. Each tool should be:
+- Single-purpose
+- JSON output
+- ≤100 LOC
+- Listed here when added
+
+| Tool | Intent | Notes |
+|---|---|---|
+| `weekly-program` | Concrete tool implementing the `weekly-program` capability | Operator implements when needed |
+| `form-check-protocol` | Concrete tool implementing the `form-check-protocol` capability | Operator implements when needed |
+| `deload-prompt` | Concrete tool implementing the `deload-prompt` capability | Operator implements when needed |
+
+## Build a new tool when you need it
+
+The agent's job includes building tools when the measurement doesn't exist. New tools should be small, JSON-output, single-purpose. Add an entry here via PR after shipping.

package/registry/families/agent-runtime-fitness-coach-ts/files/agent.json

@@ -0,0 +1,38 @@
+{
+  "name": "fitness-coach",
+  "description": "Voice-first personal fitness coach agent bundle — general strength & conditioning programming, technique cues, and deload protocols. Explicitly NOT a licensed physical therapist, not a doctor, not a registered dietitian. Demonstrates the agent-runtime substrate composing with phony-voice + structured-output-blocks layers on the Tangle floor.",
+  "version": "0.1.0",
+  "tags": [
+    "agent-runtime",
+    "fitness-coach",
+    "personal-trainer",
+    "voice-first",
+    "cloudflare-worker"
+  ],
+  "prompt": {
+    "systemPromptFile": "AGENTS.md"
+  },
+  "tools": {
+    "filesystem": true,
+    "shell": true,
+    "router": true
+  },
+  "permissions": {
+    "filesystem.write": "ask",
+    "shell.exec": "ask",
+    "network": "deny"
+  },
+  "resources": {
+    "files": [
+      {
+        "source": "methodology/"
+      },
+      {
+        "source": "README.md"
+      },
+      {
+        "source": "TOOLS.md"
+      }
+    ]
+  }
+}

package/registry/families/agent-runtime-fitness-coach-ts/files/methodology/deload-prompt.md

@@ -0,0 +1,106 @@
+# Deload Prompt — Methodology
+
+A deload is not a punishment for being weak; it's the recovery half
+of the stress-recovery-adaptation cycle. Programs that don't include
+them eventually stall, regress, or break the lifter.
+
+## Recognition signals — when the user needs a deload
+
+Most intermediate-and-up lifters earn a deload every 4–6 weeks of
+hard training. Outside that schedule, watch for signal clusters
+(any 2+ for 5–7 days):
+
+1. **Three or more failed working sets** at weights / reps that
+   were comfortable two weeks ago.
+2. **Sleep disruption** — falling asleep fine but waking at 3–4am
+   wired, or trouble falling asleep on training nights.
+3. **Joint soreness in places that aren't usually sore** —
+   elbow tendons, knee tracks, AC joints. Not muscle DOMS.
+4. **Resting heart rate elevated** by 5+ bpm for several mornings
+   in a row, with no obvious cause (illness, dehydration, alcohol).
+5. **Motivation crash** — gym dread the user normally doesn't
+   have. Workouts feel heavy on the walk in, not just under the
+   bar.
+6. **Plateau on rep PRs**, not just weight PRs. Volume PRs are
+   usually the last to stall; when they do, it's a fatigue signal.
+7. **Loss of pump / pump that takes much longer to come on.**
+   Reflects accumulated central fatigue.
+
+One signal alone is noise. Two or three together is signal.
+
+## Deload styles — pick by what's burning
+
+### Volume cut (most common)
+
+- Cut working sets per muscle group by ~40–50% for the week.
+- Keep load on the bar at ~80–90% of the previous week's working
+  weight.
+- Keep movement quality high — same lifts, same RPE 6–7 sets, just
+  fewer of them.
+- Best when the user is fatigued from accumulated volume but their
+  joints feel fine and their nervous system isn't fried.
+
+### Intensity cut
+
+- Keep the volume (sets × reps) similar.
+- Cut load to ~60–70% of working weight, ceiling at RPE 7.
+- Best when joints are barking — high-frequency lifters often need
+  this style after a peaking block.
+- The lifter still feels like they trained; the bar just isn't
+  trying to hurt them.
+
+### Full rest week
+
+- No lifting, no high-intensity conditioning. Walks, easy bike,
+  mobility, sleep.
+- Best when the user has 4+ recognition signals firing, illness,
+  major life stress, or coming off a meet / event.
+- Counter-intuitive but: most lifters underuse this and overuse
+  the other two. A real rest week is sometimes the best
+  performance enhancer in a year.
+
+## What to keep doing during a deload
+
+- **Sleep on schedule.** Deload weeks are when the body actually
+  banks the adaptation work — protect sleep harder, not less.
+- **Eat at maintenance**, not in deficit. Underfeeding a deload
+  defeats it.
+- **Walk daily** — 7–10k steps. Recovery loves blood flow.
+- **Mobility work** the user usually skips. Hips, T-spine,
+  shoulders.
+- **Skill work** at very low load if the user is technique-driven
+  (olympic lifters, powerlifters working on a specific cue).
+
+## What to stop doing
+
+- Hitting RPE 9–10 anywhere. The point is recovery; if it feels
+  like a normal training week, it isn't a deload.
+- Adding a "fun" novel-stimulus session. Novelty is fatigue
+  too — sled pushes, hill sprints, yoke walks all count.
+- Tracking food obsessively. If the user has tendencies that way,
+  this is the week to loosen the leash.
+
+## De-deload return protocol
+
+The week after the deload is not a PR week. It's the bridge.
+
+- **Week +1** — return to ~85–90% of the pre-deload working weight
+  for the same sets × reps. Should feel easy. If it doesn't, the
+  deload wasn't deep enough; consider another short deload.
+- **Week +2** — match or slightly exceed the pre-deload best (rep
+  PR or weight PR, lifter's choice).
+- **Week +3 onward** — resume normal progression.
+
+The most common deload mistake is rushing back to PR territory in
+week +1. Adaptation isn't fully on the books yet; the body has
+banked recovery, not super-compensation.
+
+## Output
+
+Emit the deload as a `:::artifact` block with:
+
+- Recognition signals observed.
+- Style chosen (volume cut / intensity cut / full rest) and why.
+- Adjusted week's sessions in the same shape as the user's normal
+  program (so it's executable, not abstract).
+- The de-deload return prescription for week +1.