Build1 distinct publisher3 min readUpdated
The enforcement is a bootstrap prompt the agent checks before every task, so its durability depends on per-harness plumbing. On Hermes Agent, compaction can lose it.
The Engineer · Build desk

Compiled by The EngineerSomething wrong?How this is made
The enforcement mechanism is a prompt. The differentiator, by the dev.to quickstart's own account, is a bootstrap instruction that makes the agent check for relevant skills before it starts any task, so the workflow activates without anyone typing a slash command [4]. That is a real improvement on a skill file whose invocation depends on a human remembering it while a deadline closes [2]. It is also a guarantee that lives in the context window, and the install notes say so plainly: Hermes Agent has no post-compaction hook, so a long session that compacts over its first turn can lose the bootstrap, and the documented remedy is to start a fresh session [11]. Pi approaches the same risk from the other side, with a small extension that injects the `using-superpowers` bootstrap at session start and again after compaction, and which does not need Pi's compatibility Skill tool because Pi supports skills natively [10]. One methodology, two different odds of it being in force on turn forty.
Then the portability accounting. Twelve harnesses have manifests [5][6], and there is no global install; you install per agent [7]. OpenCode counts as a separate path even on a machine where another harness is already set up [12], Claude Code has two marketplaces to pick between, Anthropic's official one and obra/superpowers-marketplace [8], and Cursor users find it by searching the plugin UI [20]. Portability means the methodology survives a tool switch, not that one install covers your desk.
Worth noting from the same install section: Antigravity gets its own note, including that it runs the session-start hook automatically so the skills are live from the first message [9], but Antigravity does not appear in the twelve-name manifest list [18]. Of the twelve that are listed, three carry harness-specific caveats [19]. The supported surface is already moving faster than the list that describes it.
The success signal is conversational. The guide's advice is to ask the agent what skills are available rather than trust a quiet install message, and to look for `brainstorming`, `writing-plans`, `test-driven-development` and `subagent-driven-development` in the reply [13]. When a skill is present but never fires, the guide puts the fault in the routing description rather than the model [14]. So the debugging surface is prose, and nothing here returns an exit code you could wire into CI.
What the package actually relocates is where the discipline is stored: out of the operator's memory and into the harness startup sequence [4]. Seven skills are documented as the basic workflow [17], and "write tests first, always" is one of four stated principles the agent is instructed to check [15]. That is a better home for the rule than a README, and it is still not a lock.
Follow any of these and your For You feed starts watching them — no settings page required.
Ranked by verification strength, evidence, and original report placement.
Superpowers packages a full spec-driven methodology into installable Claude Skills, enforcing brainstorming, planning, subagent-driven implementation and strict TDD instead of leaving that structure up to the user.
Superpowers is a skills package built by Jesse Vincent and the team at Prime Radiant that treats brainstorming, planning, subagent review and red-green-refactor TDD as mandatory steps the agent checks before every task rather than optional suggestions.
The stated differentiator is a bootstrap instruction that forces the agent to check for relevant skills before starting any task, so the workflow activates automatically instead of requiring the user to remember a slash command.
Superpowers ships plugin manifests for Claude Code, Cursor, Codex, Gemini CLI, GitHub Copilot CLI, Grok Build CLI, Kimi Code, OpenCode, Pi, Devin CLI, Factory Droid and Hermes Agent, so the same methodology follows the user across harnesses instead of living in one repo's .claude/skills/ folder.
Installation is harness-specific and the project is explicit that it must be installed separately for each agent used; there is no single global install.
For Claude Code, Superpowers is distributed through two marketplaces: the official Anthropic plugin marketplace and the dedicated obra/superpowers-marketplace, which also carries a few related plugins.
Evidence-backed comparisons of source perspectives and observed adoption signals. Read the methodology
Which Builder, Operator, and Investor concerns the observed source mix emphasized—not a truth score.
Evidence, demonstrated adoption, hype gap, incentives, and confidence are assessed independently, each on its own current evidence. How these are measured.
Single-source walkthrough of documented mechanics
One publisher, one article. The mechanical claims are specific and self-consistent — named harnesses, exact install commands, a verification prompt, a seven-skill workflow, and named per-harness caveats — which is more than a press-release restatement. But everything traces to one secondary guide restating project docs, with no independent test, no second publisher, and no artifact-level verification of the twelve manifests, so the evidence base is narrow.
Availability documented, adoption unmeasured
The supplied material establishes only that Superpowers is packaged and installable — two Claude Code marketplaces, a Cursor marketplace listing, manifests for twelve harnesses. There are no install counts, download figures, named teams, deployment disclosures or usage telemetry, and distribution breadth cannot substitute for uptake. No adoption value is assigned.
Enforcement framing outruns the plumbing
The claim is that spec-driven work becomes a precondition and that mandatory steps cannot be skipped under pressure. The same source shows the precondition is a bootstrap instruction the agent is asked to consult, whose persistence depends on per-harness hooks — Pi needs an extension to reinject it, Hermes Agent can lose it entirely after compaction, and installation is per harness with no global path. Combined with zero outcome or adoption data, the enforcement language is moderately overstated relative to what is evidenced; the gap is not larger because the source itself discloses the caveats plainly.
No disclosed relationships or commercial terms
The supplied source names the project's authors (Jesse Vincent and Prime Radiant) and its distribution channels but discloses nothing about pricing, licensing, funding, sponsorship, or any relationship between the guide's author and the project. With one publisher and no disclosure, any incentive reading would be invented rather than sourced.
Mechanics credible, effects and uptake unknown
Confidence is moderate-low. What the package is, how it installs, how to verify it, and where enforcement is fragile are described specifically enough to act on, and the source undercuts its own headline by documenting the Hermes Agent compaction risk — a mark of internal candor. But there is one publisher, no independent replication, no adoption measurement, and no incentive disclosure, so claims about durability and about the methodology's effect on outcomes remain unverified.
build
Claude Code now outruns Copilot roughly two to one in JetBrains' survey of 15,000 developers1 distinct publisher
build
Claude Code's agent-team panes need tmux, and Anthropic says Windows Terminal is out1 distinct publisher
build
Per-developer environments hit their ceiling the day one engineer ran five agents1 distinct publisher
science
OX Security says MCP command execution is a design choice, so server owners own the risk1 distinct publisher
Distinct publishers with included, body-backed reporting in this cluster.
dev.to
1 article · August 24, 2026