Library
Every seed pack in the library, grouped by what it is for. A pack is a collection of nouns, the reusable things, and verbs, the work an authorized worker can claim.
Every pack is a proposal: checked offline, but not installed, admitted or run by any worker yet.
The library in numbers
- 59 packs
- 1,177 nouns
- 381 verb templates
- 37 worked examples the library recomputes
- 10 packs with code and tests
Counted from the library at commit d837b30 ·
Every noun and verb was checked offline with Tessera’s own models. Read the validation receipt
Code + tests The pack ships its own code with tests, which the library’s CI runs for that pack.
Interactive Task & Task-Object ExplorerAgent Harness Research FeedsMeta Hackathon 2026 Ecosystem ExplorerThe library on GitHubSubmit a seed pack
Interactive Task & Task-Object Explorer
Tasks and Task Objects -- on the wire, verbs are tasks and nouns are task objects. Search and filter proposed Seed Packs, atomic Tasks, reusable Task Objects, and decomposed Meta Hackathon 2026 repositories right in your browser with zero network calls.
Honesty boundary: every seed pack, task template, task object, and hackathon decomposition below is an offline proposal or static public-source mapping -- none is claimed as an installed runtime service or reproduced upstream benchmark.
Enable JavaScript to use instant multi-faceted search across Tasks, Task Objects, and the Meta Hackathon 2026 ecosystem, or browse every seed pack by category below.
Encapsulating Feed Task Objects · HarnessTax & Routing
Agent Harness Research Feeds
An Agent Harness Research Feed is an encapsulating Feed Task Object (noun.feed_source / noun:harness.feed on the wire) that bundles dated capability profiles, reproducible trial receipts, and measured HarnessTax -- the static system-prompt and tool-schema token overhead paid on every turn -- across configurable coding harnesses. Because empirical benchmarks (arXiv 2311.10054, 2512.05858, 2508.19764, 2602.12285, 2402.10962) show expert role-play adds zero factual accuracy while irrelevant or demographic persona fluff degrades agent performance by 26-30% and instructions drift within 8 turns, ta-amps separates routing descriptions from permissions ('hierarchy routes, it never authorizes') and puts precision into Task Objects, output contracts, scoped tool limits, and context shims.
Status: Live daily service checks run for our own endpoints today; Herma ships PROV-HARNESS-FEEDS-GUIDE for zero-weight context shimming, while automated external GitHub/paper feed collectors remain proposed.
Harness Square
Comparative evaluation & HarnessTax baseline
Tracks structured harness profiles, static prompt/tool-schema token overhead (HarnessTax), and paired trial receipts under pinned task conditions.
Muse Code
Native Meta coding harness (X-Herma-Harness: muse-code)
Verified end-to-end claiming leased Tasks on the shared Tessera queue and receiving caller-scoped Herma context shims over /v1/responses and /v1/chat/completions.
OpenCode
Configurable open-source terminal & swarm harness
Completed native scoped task lifecycles on Tessera and scales horizontally across up to 256 concurrent workers over sharded queues (q01..q16) with inline Herma routing.
Pi
Multi-process swarm & lightweight worker harness
Completed native worker lifecycles and multi-process concurrency runs against leased Tessera queues with fenced SHA-256 receipts (verifier != builder).
Horizontal Swarm + 4-Tier Vertical Compute Cascade
Up to 256 concurrent OpenCode and multi-harness workers shard leased Tasks across Tessera queues (q01..q16, SQLite/Durable Object shards, CAS blobs) with independent verification (verifier != builder), turning verified traces and rejected-attempt tombstones (DPO pairs) into training data for a four-tier cascade:
Tier 0 · Deterministic Herma Rules (< 2 ms)
Capability firewall, volatile facts, and JSON repair
Executes inline at zero LLM token spend; deterministic repair alone lifted unchanged Llama 3.2 3B from 52.0% to 64.7% on our synthetic holdout.
Tier 1 · Local Llama 3.2 1B & 3B Modular PEFT LoRAs
Hot-swappable specialist Task Objects on a single local GPU
Routes across composable PEFT adapters (shim-router, lora-as-memory, task-decomposer, verifier-dpo) with zero per-turn HarnessTax (1B: 12.0% -> 50.7%; 3B: 52.0% -> 72.7% on synthetic holdouts).
Tier 2 · Kaggle 2xT4 Muse Glimmer 30B NF4 QLoRA
30B open-weights reasoning & LoRAs as Parametric Memory
4-bit NF4 QLoRA adapter trained on free Kaggle 2xT4 GPUs lifted tool-call accuracy from 93.3% to 99.3% (149/150) on a fresh seed-20261030 holdout.
Tier 3 · Frontier Meta Muse Spark 1.3 & Muse Wonder 1.0
Novel multi-step synthesis & cross-agent review
Reserved for frontier synthesis in Standard mode (ContributorTierRefusedError blocks contributor tiers); verified receipts distill back down into Tier 1 and Tier 2 adapters.
Nonprofit agent kits and public-solution packs 12 packs
Agent kits that a nonprofit’s own agent runs on its files. The agent drafts, and it never decides about a person or contacts anyone. Also here: how to build a new kit, and two packs that answer from official public sources, citing every point.
Show every category-
Public-good kits
Building agent kits that nonprofits can use
15 nouns 8 verbs
-
Donor thank-you letters kit
A donor thank-you letter agent kit with privacy checks
14 nouns 5 verbs
-
Flyer translation check kit
A multilingual service flyer translation check agent kit
13 nouns 4 verbs
-
Food bank inventory and pickup kit
A food bank's inventory and pickup scheduling agent kit
13 nouns 4 verbs
-
Funding impact summary kit
A grant renewal funding impact summary agent kit
13 nouns 5 verbs
-
Handwritten statements to text kit
A handwritten statements to text agent kit
16 nouns 6 verbs
-
Monthly program impact report kit
A community nonprofit's monthly program impact report agent kit
13 nouns 5 verbs
-
Policy research brief kit
A policy research brief agent kit that cites every clause
14 nouns 5 verbs
-
Victim support referral kit
A victim support referral options agent kit
13 nouns 5 verbs
-
Volunteer shift schedule kit
A disaster relief hub's volunteer shift schedule agent kit
14 nouns 5 verbs
-
Policy and law watch
Watching official public sources for new laws and rules on a topic
15 nouns 5 verbs
-
Public budget report
Reporting on a public state or city budget, one program area at a time
7 nouns 5 verbs
Video, voice and media 11 packs
Making a video with code, from script to release: scenes, motion, a synthetic voice labeled as AI, captions, editing, checks, review, cut-downs, thumbnails and an audience test. A person approves what is published.
Show every category-
Video showcase
Making a truthful demo video
15 nouns 8 verbs
-
Video and motion with code
Video and motion made with code
28 nouns 8 verbs
-
Visual and 3D rendering
Rendering images, motion graphics and 3D scenes and checking the results
11 nouns 5 verbs
-
Voice narration
Narration with a text-to-speech voice, labeled as AI
28 nouns 6 verbs
-
Narrated explainer video
Making a narrated explainer video with code, every step a verb with a receipt
26 nouns 9 verbs
-
Video editing
Editing a video from its script, with every revision kept and a person locking the cut
31 nouns 9 verbs
-
Video QA checks
Automated checks on an encoded video beyond the eight post-encode checks
20 nouns 7 verbs
-
Video review
Reviewing a video with a fixed rubric, run by a person or a worker that did not make it
26 nouns 7 verbs
-
Social cut-downs
Deriving platform cut-downs from a longer video, approved by a person before anyone posts them
25 nouns 5 verbs
-
Thumbnails and titles
Thumbnail and title candidates, checked for small sizes and honesty, tested by plan and picked by a person
24 nouns 6 verbs
-
Video audience acceptance
Testing whether a video's intended viewers understand it, with pass criteria fixed before any session
16 nouns 5 verbs
Research and analysis 7 packs
Questions answered from evidence: cited research reports, public companies and options chains described but never as financial advice, checked CSV analysis, repository reviews, measured cost savings, and synthetic cases screened for exploitation without accusing anyone.
Show every category-
Deep research report
Answering a research question with a cited report that states its limits
29 nouns 9 verbs
-
Public company research
Researching public companies from official sources, never as financial advice
27 nouns 8 verbs
-
Options chain review
Reviewing an options chain descriptively, never as financial advice
27 nouns 7 verbs
-
Checked data analysis
Analyzing a CSV file with a check at every step
32 nouns 8 verbs
-
Repository review
Reviewing a code repository with evidence
66 nouns 6 verbs
-
Cost optimization
Finding and measuring cost savings
8 nouns 6 verbs
-
Exploitation evidence
Spotting exploitation and scams in synthetic cases, never accusing anyone
70 nouns 11 verbs
Quality, testing and accessibility 8 packs
Checking work the way its users would: an accessibility audit against WCAG 2.2 AA, acceptance tests with criteria agreed first, usability and design reviews, exploratory sessions, a launch check, a local-only security review and clearer docs. Every sign-off and go decision stays with a person.
Show every category-
Accessibility audit
Auditing a page or app against WCAG 2.2 level AA
23 nouns 13 verbs
-
User acceptance testing
User acceptance testing with a person's sign-off
21 nouns 7 verbs
-
Usability heuristics review
Reviewing a page or app against ten usability heuristics
25 nouns 5 verbs
-
Screenshot design review
Reviewing a user interface screenshot against design rules
28 nouns 6 verbs
-
Exploratory testing charters
Exploratory testing in time-boxed sessions
15 nouns 4 verbs
-
Launch readiness
Deciding whether a site or app is ready to launch
18 nouns 8 verbs
-
Security review
Authorized, local-only security review of code
11 nouns 6 verbs
-
Docs improvement
Making documentation clearer for a named reader
8 nouns 5 verbs
Engineering workflow 11 packs
Work that comes back in every project: direction turned into atomic verbs, big jobs split into steps, unattended runs that leave a record, status with evidence, findings fixed and verified, releases and renames, reuse before rebuilding, file triage and claims that say how they are known.
Show every category-
Direction to work
Turning a stakeholder's message into direction nouns, finding nouns and atomic verbs
17 nouns 7 verbs
-
Decomposition patterns
Breaking work into verbs with reusable patterns
24 nouns 4 verbs
-
Atomic step contract
Giving each atomic step exactly what it needs
17 nouns 7 verbs
-
Unattended run record
Running work unattended and leaving a record anyone can check
17 nouns 7 verbs
-
Status and handoff
Answering what is done, what is next and where to resume, with evidence
15 nouns 6 verbs
-
Findings to fixes
Turning every noticed error, issue, design flaw or annoyance into recorded, fixed and verified work
24 nouns 10 verbs
-
Release and verify
Releasing a change and proving what is live
16 nouns 7 verbs
-
Rename and relink
Renaming products, repositories and pages and rewriting every reference to the new names
17 nouns 7 verbs
-
Retrieval triggers and reuse
Looking up existing nouns before generating and reusing past solutions safely
18 nouns 7 verbs
-
File triage at scale
Organizing very large collections of files without losing any
18 nouns 9 verbs
-
Measured claims
Technical notes whose every claim says how it is known
12 nouns 4 verbs
Learning and evaluation 5 packs
Measuring before trusting: a small open model tuned round after round and kept only when a rule fixed in advance says so, fair evaluations with pass marks set first, checks proven on good and bad examples, current benchmark scores and a person’s recorded choices.
Show every category-
Continuous tuning
Tuning a small open model round after round, kept only when a rule fixed in advance says so
33 nouns 9 verbs
-
Evaluation design
Fair, pre-registered evaluations
10 nouns 7 verbs
-
Checks of checks
Proving a test or rubric works before trusting it
8 nouns 4 verbs
-
Benchmark scores
Keeping benchmark scores current and comparable
10 nouns 5 verbs
-
Taste palettes
Comparing options side by side and keeping a person's choice
8 nouns 5 verbs
Architecture and discovery 5 packs
Finding what exists before building, then fitting the pieces together: new tools and repositories trialed, prior art surveyed, stale knowledge flagged, tested code composed without rewriting it, and any agent harness connected and proven end to end.
Show every category-
Tool discovery
Discovering, trialing and deciding on new tools
23 nouns 7 verbs
-
Prior art and alternatives
Finding what already exists and what the alternatives are, before anything new is built
21 nouns 7 verbs
-
Freshness checks
Finding stale, deprecated or end-of-life knowledge
7 nouns 4 verbs
-
Code composition
Composing verified code without rewriting it
35 nouns 7 verbs
-
Harness onboarding
Connecting an agent harness and proving its lifecycle
9 nouns 5 verbs