These skills do not belong to one lifecycle phase. They run before engagement, across phases, or as standalone specialized operations: plan intake, large-effort mapping, stall diagnosis, documentation and writing aids, reporting and dashboards, benchmarking, design extraction, and server hardening.

For the full arc across phases, see The skill workflow. For the alphabetical list, see Skill index.

---

Evaluate an incoming project plan against the principles and conventions, and produce a READY or NOT READY verdict.

• What it does: scores the plan through three lenses (principles alignment, conventions completeness, and a pre-flight of the build commands), closes gaps conversationally, and writes a plan audit report with a verdict.
• When to use it: when a new project arrives, when adapting a foreign plan, or before seed-conventions on an unfamiliar codebase.
• Hard gate: do not start build skills until the verdict is READY. A plan missing test commands, scope boundaries, or success criteria produces drift.
• Handoff: survey-context (READY), seed-conventions (needs bootstrapping), elaborate-spec (needs elaboration), migrate-spec (foreign format), or grill-me.

Plan a large effort as a shared map of decision tickets on an issue tracker, resolved one at a time.

• What it does: charts a wayfinder map (a single tracker issue) with child decision tickets (research, prototype, grilling, task), works them one at a time until the route is clear, and records each decision back on the map. It plans, it does not build.
• When to use it: when an idea is too big for one session, needs structured exploration before implementation, or requires mapping decisions before building.
• Key rule: never resolve more than one ticket per session (except research). Refer to every map and ticket by name.
• Note: this skill deliberately uses the issue tracker for a decision map, which is a scoped exception to the general convention that a skill does not create tracker issues.

Diagnose why agent orchestration stopped producing progress.

• What it does: reads the state, checks for contention, inspects the background terminals, classifies the stall type (waiting_approval, blocked_dependency, agent_exhausted, misconfigured_loop, external_io, unknown), and recommends one recovery action.
• When to use it: when work appears hung, there is no output for several minutes, or a subagent never returned. It is invoked by /loop, dispatch-agents, and execute-group.
• Hard gate: do not restart work blindly. Run the diagnostic first.
• Handoff: survey-context if the state is unclear, or resume the prior skill.

---

Keep the concept-wiki bundle consistent with its source docs.

• What it does: three operations. Ingest (regenerate concept pages from a source doc, a SKILL.md, or the conventions), lint (stale, orphan, missing cross-reference, contradiction, broken link), and query (search across the concept pages).
• When to use it: to keep the wiki current with the skills, the conventions, and the agent guide.
• Hard gate: none stated. The verify confirms no concept page is older than its source.

Write or rewrite technical text with the rules of ASD-STE100 Simplified Technical English.

• What it does: selects a mode (pragmatic or strict), classifies each passage as procedural or descriptive, fixes the vocabulary, applies the core limits (sentence length, approved modals, active voice), and runs a deterministic lint gate. It enforces the 53 rules of the standard.
• When to use it: for documentation, a README, a runbook, an error message, release notes, or on "de-slop" and "make this readable".
• Hard gates: classify the text first. Never change code, identifiers, CLI flags, or quoted errors. Never claim STE compliance; final approval rests with the writer.

---

Start a browser-based, read-only dashboard that visualizes the project status from the cockpit.

• What it does: starts a local server that reads the cockpit files (state, release plan, execution status) and serves a read-only PM view and a JSON status endpoint. It prefers the truenorth:// resources when driven from the MCP server.
• When to use it: to visualize the architecture, the plans, and the status.
• Hard gate: the dashboard is read-only. Do not use a visualization to make a decision without consulting the source data.

Run deterministic skill benchmarks with isolated with-skill and without-skill contexts.

• What it does: executes a versioned JSON definition in fresh disposable workspaces, runs direct code graders, and emits one stable JSON schema with per-run evidence and weighted train and validation deltas.
• Command: node skills/run-benchmark/scripts/run.js --definition <path> --skill skills/<name>/SKILL.md --output .agent/tasks/<capsule>/benchmark-<name>.json.
• Baseline: use --update-baseline <path> before a change and --check-baseline <path> after it. A lower validation delta fails.
• Hard gate: missing definitions, skills, or baselines are unavailable and non-passing. Failed or timed-out graders remain failed evidence. Scores measure change; they do not certify correctness.

---

Fetch current library docs through the Context7 MCP server instead of training data.

• What it does: resolves a library id, queries the docs (one concept per call), caches the result with a TTL, and answers from the fetched docs with a citation. It bounds itself to three calls per question.
• When to use it: on a setup, API, or code-example question for a specific library.
• Hard gates: at most three Context7 calls per question. Check the cache before an HTTP fetch. On a quota or rate-limit error, emit a CONTEXT7_UNAVAILABLE block rather than silently answering from training data.

Extract a DESIGN.md from an HTML prototype using a headless browser.

• What it does: launches a headless browser in a light and dark dual pass, collects the computed styles, classifies tokens (colors, typography, spacing, rounding, components), generates the design prose, and validates with a design-token linter.
• When to use it: when the user has an HTML prototype and wants a DESIGN.md, or right after a new project is scaffolded.
• Hard gates: the extraction must use a headless-browser dual pass; static-HTML tokens are invalid. Flag a low-confidence assertion. Do not ship without running the linter.
• Handoff: grill-me with the uncertain-decisions context.