agentsclimarketplace

Esat frontier

Skill Renn-Labs/eskill-analyze/skills/esat-frontier

Frontier-led world-class level-up analysis. Reuses eskill-analyze's engine, then runs a configurable model-fusion panel where a frontier lead such as Fable synthesizes independent reviewers such as Sonnet 5, Codex Medium, Grok, and optional fleet/OpenRouter voices. Use for /esat-frontier, frontier analysis, Fable-led analysis, model-fusion review, or highest-stakes analysis where one lead frontier model should judge, reconcile, and route a configurable council rather than relying on the fixed esat trio.From its SKILL.md

Install
npx -y skills add Renn-Labs/eskill-analyze --skill esat-frontier

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 1 stars1 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

5.3 KB, as published. Nobody here has run it

eSkill: Analyze Frontier — Frontier-Led Fusion Analyzer

esat-frontier = eskill-analyze + a frontier-led configurable fusion panel. It keeps the original analysis engine intact, then replaces the fixed trio stress test with a lead frontier model and a configurable roster of independent reviewers.

Default posture:

  • Lead: Fable if the harness exposes it; otherwise the strongest available frontier model.
  • Reviewers: Sonnet 5, Codex Medium, Grok, plus optional fleet/OpenRouter voices when configured.
  • Synthesis: the lead does not average votes. It verifies claims, reconciles disagreements, and emits one evidence-backed verdict.

Shared routing contract (load first)

Before Phase 9 dispatch, read and obey:

${CLAUDE_SKILL_DIR}/../eskill-common/references/model-routing.md

That contract defines semantic role vs requested vs actual route, preference precedence, harness identity, strict pins, disclosed automatic fallback, strongest-restriction policy, metered consent, two-stage manifest, and two-phase deduplication. This skill owns Phase-9 composition only.

Engine — reuse eskill-analyze verbatim

Run the full eskill-analyze protocol for Input Parameters, Triage, Delegation, Framework Selection, Evidence Gathering, and Steps 1-8. Read and follow, in order:

  1. ${CLAUDE_SKILL_DIR}/../eskill-analyze/SKILL.md
  2. ${CLAUDE_SKILL_DIR}/../eskill-analyze/references/triage-guide.md
  3. ${CLAUDE_SKILL_DIR}/../eskill-analyze/references/mental-models.md
  4. ${CLAUDE_SKILL_DIR}/../eskill-analyze/references/analysis-protocol.md — run Phase 0 + Steps 1-8 + Phase P Premortem (when triage activates it). Skip that file's Phase 9.
  5. ${CLAUDE_SKILL_DIR}/../eskill-analyze/references/world-class-signals.md

Frontier overrides

  • Phase 9 is mandatory for standard and deep; for quick, run a compact two-voice panel unless the user explicitly asks for no panel.
  • Phase 9 uses this skill's fusion panel, not eskill-analyze's single critic and not esat's fixed trio.
  • The lead is responsible for judgment, not just summarization. It must verify claims against the item/repo before adopting reviewer output.
  • Model labels are portable profiles, not provider IDs. Resolve Fable, Sonnet 5, Codex Medium, and Grok through references/model-profiles.md before dispatching reviewers.
  • Preserve the current host model as host; managed reviewer roles resolve independently and must not claim the host was hot-swapped.
  • Codex and Grok dispatch independently (peer codex / peer grok). Never use roster-driven peer trio for this tier.
  • Exact pins are strict; unavailable routes are reported, not silently substituted. Disclose actual model or account default — never stale fixed version labels.

Phase 9 — Frontier Fusion Panel

After Steps 1-8 and Phase P (when activated) produce the draft analysis, read and execute:

  1. ${CLAUDE_SKILL_DIR}/../eskill-common/references/model-routing.md (if not already loaded)
  2. ${CLAUDE_SKILL_DIR}/references/model-profiles.md
  3. ${CLAUDE_SKILL_DIR}/references/frontier-panel.md

Configuration

Use environment variables when the harness or shell supports them (legacy environment tier in precedence; explicit invocation and session outrank them):

VariableDefaultEffect
ESAT_FRONTIER_LEADfablePreferred lead profile. Resolve aliases through model-profiles.md; if unavailable, use the strongest available frontier profile and state the fallback.
ESAT_FRONTIER_ROSTERsonnet-5,codex-medium,grokComma-separated reviewer profiles. Resolve aliases through model-profiles.md; unknown or unavailable labels are skipped with a panel note.
ESAT_FRONTIER_FLEET01 adds the fleet OSS swarm leg through fleet-fuse when sensitivity, redaction, permission, budget, and metered consent allow.
ESAT_FRONTIER_SENSITIVITYmediumhigh keeps review local/first-party only; medium allows redacted external reviewers; low allows broader low-tier pools; invalid fails closed.
ESAT_FRONTIER_BUDGET_USDunsetOptional positive numeric cap for external fleet calls. Absent → disclose provider/account cap only. Invalid → fail closed.

Preference precedence (from the shared contract): explicit invocation > session > project > user > legacy environment > harness > portable default.

Output

Use ${CLAUDE_SKILL_DIR}/../eskill-analyze/assets/output-template.md plus the ### Frontier Fusion Panel block defined in references/frontier-panel.md.

Set the triage header to esat-frontier (frontier fusion) and save output to .omc/plans/esat-frontier-{date}-{focus}.md.

Post-Analysis Routing

Same as eskill-analyze: route confirmed action lists to sprint or overnight execution only after the user confirms the actions.

Keep looking

Skills are one crate of 608,666. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.