HELM's system designer. Sees the structure before the code — protocols, service boundaries, data flow, failure modes — and checks the design can hold the weight the founding bet will put on it. Auto-invoked by the Orchestrator at the Plan gate when assigned; never summoned by the MD directly.
HELM's research-domain structure reviewer. Sees the argument before the prose — thesis, claim hierarchy, evidence flow, and failure modes — and checks the argument structure is deliberate and the thesis is supportable by planned evidence. Auto-invoked by the Orchestrator at the Plan gate when assigned; never summoned…
HELM's compliance reviewer. Audits business plans, strategy docs, and product specs for external claims, regulatory obligations, contractual commitments, and liability exposure. Holds that every claim is defensible and every commitment is deliverable. Auto-invoked by the Orchestrator at the Plan and Build gates when…
HELM's standing adversarial reviewer. Argues against the project's dominant assumption, challenges overconfident claims (the high-confidence check at confidence ≥9), and second-passes verified work for overconfidence. Auto-invoked by the Orchestrator at the Spec and Verify gates — the MD never summons it directly.
HELM's status renderer — turns the local ledgers into a single self-contained HTML page showing project status, phase, cost, cache savings, and lessons. No server, no daemon, no network. Use when the MD asks to "see" the project or portfolio, wants a status page, or asks where the tokens went. Also offered at the Ship…
HELM's implementer. The only role that writes the work — features, then the fixes other specialists' findings generate. Builds exactly the routed task, surfaces gaps instead of filling them, and exits on cited build/test evidence, never "it compiles." Routed by the Orchestrator during the Build phase; never summoned…
HELM's numbers reviewer. Adversarial about assumptions and figures — hunts the load-bearing assumption, the figure that doesn't tie out, the untested sensitivity. Built ≠ validated; a number with a source ≠ a number that holds. Auto-invoked by the Orchestrator at the Build gate when assigned; never summoned by the MD…
HELM's fact-checker. Adversarial about accuracy — the load-bearing claim, the surprising statistic, the misremembered date. Holds that cited ≠ true. Auto-invoked by the Orchestrator at the Build gate when assigned; never summoned by the MD directly.
Route each task to the cheapest model that can still do it, across any mix of local and cloud models you choose. Bring your own models via an OpenAI-compatible endpoint, declare their cost and capability, and the CEO scores each task and sends it to the cheapest capable tier — escalating to a stronger model only when…
Onboard a new project the HELM way — describe it, review what's in the folder, recommend a project-specific founding bet, and assemble a deterministic, project-specific team from the role library. Use at the start of any new project, or when the user says "onboard", "set up HELM", "start a project", or "build me a…
HELM's standing operating rule — four enforced defaults on cost, focus, portfolio discipline, and completion honesty that apply to the Orchestrator and every agent for the whole session. Model routing (the orchestrating model never does grunt work), tangent & scope guards (every task traces to the founding bet or…
HELM's business deliverable builder. Routed during Build phase to produce the spec, strategy doc, market analysis, product plan, or financial model — every figure sourced and every assumption labelled. Reviewers find gaps; the Operator fills them. Surfaces unsupported assumptions as findings, never invents numbers.…
The HELM CEO/Orchestrator. Coordinates a project team through the lifecycle (spec → plan → build → verify → review → ship), routes work to specialists, and AUTONOMOUSLY invokes the right specialist when a phase hits its mechanical completeness signal. The single point of contact for the human (the MD). Use to drive an…
The HELM Portfolio CEO — the exec view across ALL projects, products, and open topics in a portfolio. Discovers nested projects, derives each one's status from mechanical signals, and briefs the MD on what needs them now and what's at risk across the portfolio. Use at a portfolio root to get a status briefing, spin up…
HELM's founding-bet guardian. Measures every task, feature, and "while we're here" addition against the one outcome the project is wagering on, and cuts what doesn't serve it. Auto-invoked by the Orchestrator at the Plan and Review gates; never summoned by the MD directly.
HELM's correctness reviewer. Adversarial about behaviour — walks the unhappy path, the empty state, the corrupted input, not the demo. Holds that a clean build is not a verified one. Auto-invoked by the Orchestrator at the Build gate when assigned; never summoned by the MD directly.
HELM's cold-reader advocate. Walks any written work as an intelligent stranger with no author context and names every place comprehension breaks — undefined jargon, unexplained leaps, buried thesis, unsupported claims. Auto-invoked by the Orchestrator at the Review gate when assigned; never summoned by the MD directly.
HELM's meta-router. Maps a lifecycle phase or an ad-hoc request to the specialist that should handle it, using mechanical signals from repo state. Used by the Orchestrator to decide who to invoke next. Not a decision-maker — a lookup.
HELM's Scribe. The write-side of the lessons ledger — distills every resolved failure (CHALLENGE/REJECT/FAIL verdicts, escalations) into a structured lesson in .helm/lessons.jsonl, and produces the ship retro. Auto-invoked by the Orchestrator when an adverse verdict is resolved and at the Ship gate; never summoned by…
HELM's security reviewer. Audits anything that touches networks, file transfer, external systems, authentication, or sensitive data. Holds the standard that transfers and inputs are encrypted, authenticated, and integrity-checked. Auto-invoked by the Orchestrator at the Plan and Build gates when assigned to the team …
HELM's sources reviewer. Audits research claims, citations, and factual sourcing. Holds that every load-bearing claim traces to a credible, correctly-represented source — no unsourced assertions, no misquoted or hallucinated references, no plagiarism. Auto-invoked by the Orchestrator at the Plan and Build gates when…
HELM's business-domain audience advocate. Reads any deliverable that asks a stakeholder/investor/board to decide as that decision-maker would—finding where the ask is unclear, value is buried, or the decision they're being asked to make isn't stated. Auto-invoked by the Orchestrator at the Review gate when assigned…
HELM's strategic structuralist. Sees the business plan's spine before the build — thesis, dependent bets, sequencing, and failure conditions — and checks the strategy can hold the weight the founding bet will put on it. Auto-invoked by the Orchestrator at the Plan gate when assigned; never summoned by the MD directly.
HELM's startup update check. Mechanically compares the installed HELM version against the latest published on GitHub and, if behind, recommends the one-line update command to the MD. Runs once per session, network-optional (skips silently offline), and never modifies the install itself. Invoked first by the…
★not rated 6 1mo agoA78 tokens
originalMIT
At most 3 mods per repository are shown here, and a mod shipped inside a plugin is left to that plugin's page — the rest are on their repository pages: