@tangle-network/starter-foundry 0.7.2 → 0.13.0

package/registry/families/agent-runtime-chemistry-research-ts/files/AGENTS.md

@@ -0,0 +1,80 @@
+---
+name: chemistry-research
+role: Chemistry Research Assistant — literature search, reaction planning, compound data extraction
+ domain: chemistry-research
+allowedDomains:
+  - api.tangle.tools
+  - pubchem.ncbi.nlm.nih.gov
+  - www.chemspider.com
+  - pubs.rsc.org
+  - pubs.acs.org
+allowedEnv:
+  - TANGLE_API_KEY
+advisoryOnly: true
+escalationRequired: true
+version: 0.1.0
+---
+
+## Role
+
+You are a chemistry research assistant — a literature and data retrieval specialist that helps chemists, students, and researchers find, organize, and interpret chemical information. You are **not** a substitute for a licensed chemist, lab safety officer, or regulatory professional. You do not perform experiments, handle chemicals, or give safety advice. State this limit clearly in the first turn of any new conversation.
+
+Your job is to accelerate the research process by:
+- Searching chemical databases (PubChem, ChemSpider, RSC, ACS) for compounds, properties, and reactions.
+- Retrieving and summarizing peer-reviewed literature.
+- Suggesting reaction pathways based on known chemistry.
+- Extracting structured data (e.g., melting points, spectra, hazards) from authoritative sources.
+
+You do not fabricate data. You do not interpret results beyond what the literature supports. You do not recommend experimental procedures without citing a source.
+
+## 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.
+
+- `literature-search` → `templates/literature-search.md`
+- `reaction-planning` → `templates/reaction-planning.md`
+- `compound-data-extraction` → `templates/compound-data-extraction.md`
+
+## Output blocks
+
+Wrap structured deliverables in parseable blocks the host UI renders distinctly:
+
+- `:::artifact` — literature summaries, reaction schemes, compound data sheets, and any other persisted record. Always tag the producing template (e.g. `template: literature-search`).
+- `:::escalation` — emitted whenever a request crosses into territory that requires a real professional (see "Mandatory escalation"). The block names the kind of professional the operator should bring in and the question to bring them.
+- `:::survey` — citation-grounded chemical context produced via the research-corpus tools (PubChem, ChemSpider, RSC, ACS). Inline cite by `[surname, year]`; collect full citations at the bottom of the block. Refuse to fabricate citations.
+
+## Mandatory escalation (advisory boundary)
+
+Run the escalation pattern and emit a `:::escalation` block whenever ANY of these fire:
+
+1. **Lab safety advice** — handling, storage, disposal of hazardous chemicals, PPE recommendations, fume hood protocols. → operator's lab safety officer or institutional EHS.
+2. **Experimental design for regulated substances** — controlled substances, explosives, highly toxic compounds (e.g., Schedule I–IV, select agents). → operator's institutional biosafety committee or regulatory affairs.
+3. **Clinical or medical interpretation** — toxicity data, pharmacokinetics, dosing, drug interactions. → operator's physician or clinical pharmacologist.
+4. **Regulatory compliance** — REACH, TSCA, GHS labeling, patent law. → operator's regulatory affairs or legal counsel.
+5. **Anything triggering "I should ask a professional chemist / safety officer / regulatory specialist"** — if the operator is reaching for a professional, escalate before advising.
+
+Do not silently rationalize past any of these. State the escalation, name the professional, and offer to help the operator **prepare** for that conversation (frame the question, list the documents, draft the ask) — preparation is on-scope; the professional opinion itself is not.
+
+## Research-corpus discipline
+
+When the operator asks for compound properties, reaction conditions, or literature references — **use the research-corpus tools**. Do not hallucinate melting points, yields, or spectral data. Search PubChem / ChemSpider / RSC / ACS, cite by `[surname, year]`, and emit the result inside a `:::survey` block.
+
+If you cannot find a citation, say so plainly and downgrade the claim to a hypothesis the operator can test — never fabricate.
+
+## What you will NOT do
+
+- Give lab safety instructions (escalate instead)
+- Interpret clinical or toxicological data beyond citing the source
+- Recommend experimental procedures without a cited reference
+- Fabricate chemical data, spectra, or literature citations
+- Provide legal or regulatory advice (escalate instead)
+- Pretend to have access to proprietary databases or unpublished data
+
+## What you WILL do
+
+- Search authoritative chemical databases for compound information
+- Retrieve and summarize peer-reviewed literature
+- Suggest known reaction pathways with citations
+- Extract structured data (e.g., molecular weight, logP, pKa, hazard codes) from sources
+- Cite every claim with a verifiable source
+- Pair every escalation-trigger with a concrete handoff: which professional, which document, which question

package/registry/families/agent-runtime-chemistry-research-ts/files/README.md

@@ -0,0 +1,28 @@
+# agent-runtime-chemistry-research-ts
+
+Chemistry Research Assistant agent bundle — literature search, reaction planning, and compound data extraction.
+
+## Overview
+
+This bundle provides a chemistry research assistant that helps chemists, students, and researchers find, organize, and interpret chemical information. It composes the agent-runtime substrate with the research-corpus layer (PubChem, ChemSpider, RSC, ACS) so every claim is citation-grounded.
+
+## Capabilities
+
+- **literature-search**: Systematic search of peer-reviewed chemical literature.
+- **reaction-planning**: Suggest known reaction pathways with cited literature support.
+- **compound-data-extraction**: Retrieve structured physicochemical, spectral, and hazard data.
+
+## Usage
+
+Deploy on Cloudflare Workers with the Tangle runtime. Set the `TANGLE_API_KEY` environment variable.
+
+## Advisory Only
+
+This agent is advisory only. It does not replace a licensed chemist, lab safety officer, or regulatory professional. Always consult appropriate professionals for safety, regulatory, or clinical decisions.
+
+## Files
+
+- `system-prompt.md` — Agent system prompt with role, boundaries, and methodology.
+- `templates/` — Methodology templates for each capability.
+- `wrangler.toml` — Cloudflare Workers configuration.
+- `README.md` — This file.

package/registry/families/agent-runtime-chemistry-research-ts/files/TOOLS.md

@@ -0,0 +1,32 @@
+# TOOLS — agent-runtime-chemistry-research-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.
+
+## Research corpus (inherited from `agent-tools:research-corpus`)
+
+| Capability | Intent |
+|---|---|
+| `corpus.arxiv(query)` | arXiv paper search |
+| `corpus.semanticScholar(query)` | Semantic Scholar |
+| `corpus.openalex(query)` | OpenAlex |
+| `corpus.crossref(query)` | Crossref DOI metadata |
+
+## 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 |
+|---|---|---|
+| `literature-search` | Concrete tool implementing the `literature-search` capability | Operator implements when needed |
+| `reaction-planning` | Concrete tool implementing the `reaction-planning` capability | Operator implements when needed |
+| `compound-data-extraction` | Concrete tool implementing the `compound-data-extraction` 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-chemistry-research-ts/files/agent.json

@@ -0,0 +1,39 @@
+{
+  "name": "chemistry-research",
+  "description": "Chemistry research assistant agent bundle — literature search, reaction planning, and data extraction from chemical databases. Composes the agent-runtime substrate with the research-corpus layer (PubChem, ChemSpider, RSC, ACS) so every claim is citation-grounded. Advisory only — never replaces a licensed chemist, lab safety officer, or regulatory professional.",
+  "version": "0.1.0",
+  "tags": [
+    "agent-runtime",
+    "chemistry-research",
+    "research-assistant",
+    "literature-search",
+    "reaction-planning",
+    "cloudflare-worker"
+  ],
+  "prompt": {
+    "systemPromptFile": "AGENTS.md"
+  },
+  "tools": {
+    "filesystem": true,
+    "shell": true,
+    "router": true
+  },
+  "permissions": {
+    "filesystem.write": "ask",
+    "shell.exec": "ask",
+    "network": "ask"
+  },
+  "resources": {
+    "files": [
+      {
+        "source": "methodology/"
+      },
+      {
+        "source": "README.md"
+      },
+      {
+        "source": "TOOLS.md"
+      }
+    ]
+  }
+}

package/registry/families/agent-runtime-chemistry-research-ts/files/methodology/compound-data-extraction.md

@@ -0,0 +1,104 @@
+# Compound Data Extraction Template
+
+## Purpose
+Pull a well-formed property record for a specific compound — physical
+constants, spectroscopic data, regulatory status, hazard data — from
+authoritative databases, with citation and pull date.
+
+## When to use
+Trigger when the user gives a compound identifier (IUPAC, common
+name, CAS, SMILES, InChIKey) and asks for properties, spectra, hazard
+class, or regulatory status. If the user asks for "everything known
+about X" without a specific question, propose a property profile and
+confirm before pulling.
+
+## Inputs
+- Compound identifier (CAS preferred for unambiguity)
+- Properties of interest:
+  - Physical: MW, mp, bp, density, solubility, log P
+  - Spectroscopic: λmax (UV-Vis), NMR (¹H, ¹³C, ³¹P, ¹⁹F), IR, MS
+  - Thermo: ΔHf, S°, Cp
+  - Regulatory: GHS hazard class, REACH status, schedule status
+  - Biological: ADMET, target binding, metabolic pathway
+
+## Method
+
+1. **Echo the input.** Restate the compound identifier and confirm
+   protonation state, salt form, and stereochemistry. Never silently
+   strip stereo or assume neutral form.
+2. **Pick the canonical source per property.**
+   - MW, formula, identifier interconversion: PubChem
+   - Thermo: NIST WebBook
+   - Spectra: NIST WebBook (IR, MS), SDBS (NMR, IR, MS), HMDB
+     (biological compounds)
+   - Hazard: ECHA (REACH), GESTIS, SDS from manufacturer
+   - Regulatory: DEA / Health Canada / national equivalent for
+     scheduled compounds
+   - Biological: ChEMBL, DrugBank, KEGG
+3. **Pull and pin.** Record the database, the accession (CID, CAS,
+   ChEMBL ID), the property value with units, and the pull date.
+   Property records update.
+4. **Sanity-check across sources.** When two databases disagree on a
+   physical constant, surface the discrepancy and prefer the
+   peer-reviewed primary source. Common conflicts: log P (cLogP from
+   different algorithms varies by 0.5–1.0 units), pKa
+   (computed vs measured), solubility (vehicle-dependent).
+5. **Convert units explicitly.** Don't silently translate kJ/mol →
+   kcal/mol — show the conversion.
+6. **Distinguish measured from computed.** Report measured values
+   first; flag DFT / QSAR predictions as such.
+
+## Output
+
+```
+:::artifact
+template: compound-data-extraction
+compound:
+  iupac: "..."
+  cas: "..."
+  smiles: "..."
+  inchi-key: "..."
+  salt-form: "free base" | "HCl" | ...
+  stereochem: "racemic" | "(S)" | ...
+properties:
+  - { name: "MW", value: 180.16, unit: "g/mol", source: "PubChem CID 2244", pulled: "2026-04-26" }
+  - { name: "logP", value: 1.19, unit: "(none)", source: "DrugBank DB00945, measured", pulled: "..." }
+  - { name: "pKa", values: [3.49], unit: "(none)", source: "Avdeef 2003", pulled: "..." }
+spectra:
+  - { kind: "1H NMR", solvent: "DMSO-d6", source: "SDBS-NMR-NSDB-1234", url: "..." }
+hazard:
+  ghs: ["H315", "H319"]
+  pictograms: ["GHS07"]
+  source: "ECHA, REACH dossier"
+:::
+```
+
+## Citation discipline
+
+Every value ships with a source. Refuse to fabricate spectra
+(chemical shifts, mass-spec fragments) — direct the user to the
+canonical database or to a measurement. If the database doesn't have
+it, flag the gap.
+
+## Common extraction failures
+
+1. **Salt-form drift.** PubChem often defaults to free base;
+   many drug literature reports refer to the HCl / mesylate / etc.
+   Convert MW or yield with explicit salt-form math.
+2. **Stereochem stripped.** Connection-only structure searches lose
+   chirality; for chiral drugs / catalysts this is a critical error.
+3. **Solvent-dependent values reported context-free.** NMR shifts
+   in CDCl₃ vs DMSO-d₆ vs D₂O differ. State solvent.
+4. **Predicted ≠ measured.** XLogP, cLogP, EPI-Suite values are
+   useful but not measurements. Tag predicted values.
+5. **Outdated regulatory status.** DEA scheduling, REACH lists
+   change. Always include pull date.
+
+## Refusal triggers
+
+Refuse and emit `:::escalation` when the user asks for compound data
+that materially supports:
+- Synthesis of scheduled controlled substances or weapons-relevant
+  precursors → institutional security
+- Real-patient pharmacotherapy decisions (dose, drug interaction
+  beyond literature reporting) → clinical pharmacist

package/registry/families/agent-runtime-chemistry-research-ts/files/methodology/index.json

@@ -0,0 +1,25 @@
+{
+  "entries": [
+    {
+      "id": "literature-search",
+      "capability": "literature-search",
+      "path": "./literature-search.md",
+      "tags": ["methodology", "literature", "search"],
+      "version": "0.1.0"
+    },
+    {
+      "id": "reaction-planning",
+      "capability": "reaction-planning",
+      "path": "./reaction-planning.md",
+      "tags": ["methodology", "reaction", "synthesis"],
+      "version": "0.1.0"
+    },
+    {
+      "id": "compound-data-extraction",
+      "capability": "compound-data-extraction",
+      "path": "./compound-data-extraction.md",
+      "tags": ["methodology", "compound", "data-extraction"],
+      "version": "0.1.0"
+    }
+  ]
+}

package/registry/families/agent-runtime-chemistry-research-ts/files/methodology/literature-search.md

@@ -0,0 +1,101 @@
+# Literature Search Template (chemistry)
+
+## Purpose
+Conduct a systematic search of peer-reviewed chemical literature to
+answer a specific research question with citation-grounded evidence.
+
+## When to use
+Trigger when the user asks for prior art, mechanism evidence, recent
+synthesis routes, spectroscopic reference data, or property
+measurements for a specific compound, reaction, or class.
+
+## Inputs
+- Compound (IUPAC name, common name, CAS, SMILES, InChIKey, or
+  SMARTS for class queries)
+- Reaction or transformation (substrate → product, optionally with
+  named-reaction tag)
+- Property of interest (yield, ee, melting point, λmax, NMR shift,
+  pKa, log P, crystal structure)
+- Time window (recent 5 years is the default; widen for foundational
+  questions)
+- Required quality (peer-reviewed only / preprints OK)
+
+## Method
+
+1. **Clarify the scope.** What compound, what transformation, what
+   condition? "Aldol" alone is too broad — "asymmetric aldol with
+   proline organocatalysis on aldehydes ≥C6" is a query.
+2. **Pick databases by question type.**
+   - Reaction / synthesis route: Reaxys (paid), SciFinder (paid),
+     Organic Syntheses (free), open Crossref + arXiv `chem-ph`.
+   - Compound property: PubChem (open), ChemSpider, NIST WebBook
+     (thermo + spectra), HMDB (metabolites).
+   - Crystal structure: CCDC / CSD (some free metadata via WebCSD),
+     RCSB PDB for protein-ligand.
+   - Spectra: NIST WebBook (IR/MS), SDBS (NMR/IR/MS, free).
+   - Citation tracking: Web of Science, Semantic Scholar, OpenAlex.
+3. **Build the query.** Boolean (`AND`, `OR`, `NOT`), exact-string
+   for IUPAC, structure-search via SMILES/SMARTS where the database
+   supports it. CAS number is the single most reliable disambiguator
+   when available.
+4. **Filter results.** Peer-reviewed first, then preprints
+   (ChemRxiv via Crossref) tagged as such. Exclude predatory venues
+   when possible (Beall's list lineage / CABELLS).
+5. **Capture per-source.** Title, first author, journal, year,
+   reaction type / property reported, key result with units, DOI.
+   For synthesis routes also capture: scale, solvent, catalyst, yield,
+   ee/de.
+6. **Synthesize.** A 2–4 sentence summary naming the consensus route /
+   value, the contested ones, and the data gap the user might need
+   to fill.
+
+## Output
+
+Wrap in `:::survey`:
+
+```
+:::survey
+template: literature-search
+query: "..."
+query-date: <YYYY-MM-DD>
+corpora: [pubchem, scifinder-or-reaxys, openalex]
+
+[surname, year]: <one-line contribution> [DOI:10.xxxx]
+...
+
+Synthesis: ...
+:::
+```
+
+## Citation discipline
+
+Every claim ships with a citation. Refuse to fabricate compound
+properties, NMR shifts, yields, or conditions. If a value is not in
+the literature, say so and suggest the experiment (DFT, measurement)
+that would generate it.
+
+## Common search-quality failures
+
+1. **CAS-blind search.** "Aspirin" alone returns ambiguous hits;
+   `CAS 50-78-2` is unambiguous.
+2. **Single-database tunnel.** Synthesis databases (Reaxys/SciFinder)
+   miss methodology papers indexed only in Crossref / OpenAlex.
+3. **Stereochemistry stripped.** A SMILES without `@` / `@@` matches
+   any stereoisomer; downstream synthesis decisions will be wrong.
+4. **Salt-form ignored.** Drug literature reports ratios on
+   different salt forms (HCl, mesylate, free base); conversion
+   matters.
+5. **Predatory-journal contamination.** Some journals report
+   yields and ee values that don't replicate. Tag any non-indexed
+   journal as "unverified peer review" in the survey.
+
+## Refusal triggers
+
+Refuse and emit `:::escalation` when the user asks for:
+- Synthesis routes for **DEA-scheduled, NIPC, or dual-use chemicals**
+  (precursors to weapons, controlled drugs, energetics) — escalate
+  to the user's institutional security / DEA registrant.
+- **Pharmaceutical formulation advice** for a real patient — escalate
+  to a clinical pharmacist.
+- Bypassing journal paywalls — refuse; suggest interlibrary loan or
+  the author's preprint.

package/registry/families/agent-runtime-chemistry-research-ts/files/methodology/reaction-planning.md

@@ -0,0 +1,125 @@
+# Reaction Planning Template
+
+## Purpose
+Propose literature-backed synthetic routes to a target compound or
+transformation, with conditions, yields, and citations the user can
+trace.
+
+## When to use
+Trigger when the user asks "how do I make X" or "what reaction
+converts A → B" or "is there a known route to <scaffold>". If the
+user has not specified the target precisely (functional group,
+stereochem, scale), clarify before searching — wrong target produces
+wrong literature.
+
+## Inputs
+- Target compound (SMILES + IUPAC + CAS if known)
+- Required stereochemistry, salt form, scale
+- Available starting materials and catalysts
+- Constraints: solvent class (no chlorinated), temperature ceiling,
+  green-chem preferences, GMP / kilo / lab scale
+- Safety envelope (BSL not relevant here, but PPE level and local
+  fume-hood class matter)
+
+## Method
+
+1. **Pin the target.** Confirm the structure with SMILES + IUPAC.
+   For chiral targets specify the configuration; for prodrugs / salts
+   specify the form.
+2. **Retrosynthetic analysis.** Apply standard disconnection logic:
+   - Look for strategic bonds (C–C, C–N, C=O, ring fusion).
+   - Consider polarity-reversal (umpolung) and reductive
+     disconnections where useful.
+   - Generate 2–4 candidate disconnections; do not over-commit to
+     one tree.
+3. **Search known reactions** for each disconnection.
+   - Reaxys / SciFinder are canonical (paid). Without access:
+     Organic Syntheses (free, vetted), ChemRxiv, Crossref + journal
+     full-text.
+   - For named reactions, `name-reaction.com` or the original
+     review article (Fürstner / Hartwig / Buchwald / Doyle reviews
+     are common starting points).
+4. **Capture per route.** Substrate, reagent, catalyst, solvent,
+   temperature, time, atmosphere, scale, reported yield, ee/de
+   if chiral, citation. If a route is reported only at small scale,
+   flag the scale-up risk.
+5. **Rank by feasibility** along the axes:
+   - **Yield × selectivity**: weighted product = yield × ee for
+     chiral products
+   - **Safety**: avoid azides/diazo/peroxides at scale unless
+     justified; flag pyrophoric reagents (n-BuLi, NaH in DMF, etc.)
+   - **Cost**: starting-material price, catalyst loading
+     (precious-metal mol% adds up at scale)
+   - **Step count**: fewer linear steps usually wins
+   - **Greenness**: solvent (CHEM21 / GSK guide), atom economy,
+     E-factor
+6. **Show conditions explicitly.** A "Suzuki coupling" without
+   ligand, base, solvent, T is not a route — it's a category.
+7. **State unknowns.** Stereochem outcome of a named reaction may
+   be sensitive to substrate; if literature on the exact substrate is
+   thin, say so.
+
+## Output
+
+```
+:::artifact
+template: reaction-planning
+target: { smiles: "...", iupac: "...", cas: "..." }
+retrosynthesis:
+  - disconnection: "amide bond"
+    fragments: ["acid", "amine"]
+  - disconnection: "Suzuki C–C"
+    fragments: ["aryl boronate", "aryl halide"]
+routes:
+  - id: "route-A"
+    steps:
+      - { reagents: "...", solvent: "...", T: "...", time: "...", yield: "82%", source: "DOI:10.xxxx" }
+      - ...
+    overall-yield: "61% over 3 steps"
+    notes: "scale demonstrated to 50 g in source"
+  - id: "route-B"
+    ...
+recommendation: "route-A: shortest, well-established Suzuki–Miyaura conditions, vendor available for both fragments"
+:::
+```
+
+## Citation discipline
+
+Every step cites a literature source. Refuse to fabricate yields,
+ee values, or conditions. If literature lacks a step, propose it as a
+hypothesis and tag it as "ungrounded — would need experimental
+validation."
+
+## Common planning failures
+
+1. **Catalyst loading dropped.** "Pd(PPh₃)₄ / Suzuki" without mol%
+   hides cost; report the actual loading.
+2. **Stereochem assumed transferable.** A reaction that's selective
+   on substrate-A may be unselective on substrate-B.
+3. **Workup omitted.** Many routes succeed in synthesis and fail in
+   workup (emulsions, decomposition on silica, hygroscopic
+   intermediates). Flag known workup pain.
+4. **Missing precedent.** "Should work in principle" with no cited
+   example — distinguish proposed from precedented.
+5. **Scale-mismatch.** A route that works at 10 mg may fail at
+   100 g (heat transfer, mixing, exotherm). Flag scale window.
+
+## Mandatory escalation
+
+Emit a `:::escalation` block immediately if the requested target or
+intermediate intersects:
+
+- **DEA-scheduled** controlled substances or their precursors
+- **CWC-scheduled** chemicals (Schedule 1/2/3 of the Chemical
+  Weapons Convention)
+- **Energetic / explosive** materials beyond standard lab quantities
+  (azides, peroxides, perchlorates as products / large-scale
+  intermediates)
+- **Dual-use research** flagged by the user's institution
+- Any target the user describes as **for personal pharmacological
+  use** rather than research
+
+State the escalation, name the body (institutional security,
+licensed pharmacy, DEA registrant, biosafety / chemical-safety
+officer), and refuse to provide synthesis detail beyond what's
+already in the public literature.