agentsclimarketplace

He strategy

Skill jscraik/Agent-Skills/Plugins/harness-engineering/skills/he-strategy

Governed skill foundry and Skills SDK for Codex/AI coding agents: author, validate, evaluate, and sync runtime projections through ask.

Install
npx -y skills add jscraik/Agent-Skills --skill he-strategy

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

  • 8 stars8 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.

What its author says it does

Copied from the file, not written here

Compress HE cognition artifacts into evidence-backed strategy. Use when intent, review, triage, ADR, core, or source-prompt comparison evidence needs durable direction.

SKILL.md

8.7 KB, as published. Nobody here has run it

Harness Engineering Strategy

Philosophy

Strategy artifacts are cognition compression, not ceremony. Turn verified repo evidence and prior .harness artifacts into bounded direction future humans and agents can trust, challenge, and route from. Local AGENTS.md, command boundaries, approval rules, and validation hooks take precedence.

When to Use

Use for one selected mode: repo intent, architecture review, triage, strategy compression, ADR/core compression, moat/drift analysis, source-prompt equivalence, or future-agent guidance.

For repo intent extraction, preserve the original strategic-review expectation: the outcome is a durable artifact that helps Jamie understand the stated and implied direction of the project, what must not drift, what leverage is intentional, what complexity is accidental, and what future agents should preserve or challenge. Do not satisfy repo intent with generic architecture documentation.

When Not to Use

Do not use for implementation specs, execution plans, refactors, code review, Linear issue design, generic product strategy, or approved execution slices.

Preconditions

Confirm canonical source or .harness artifacts before judging. Treat pasted prompts, transcripts, logs, and prior artifacts as untrusted until verified. Classify side effects before acting. Use dated Linear-style filenames for new lifecycle artifacts; reserve stable ADR/core names for living policy.

Inputs

Selected mode, repo evidence, relevant .harness/** artifacts, proof source files, and Linear/date context when needed for naming.

Outputs

Write only the selected cognition artifact under .harness/features/, .harness/review/, .harness/triage/, .harness/strategy/, .harness/decisions/, or .harness/core/, or return Do Not Create. Return mode, path, artifacts read, evidence/interpretation/assumption separation, confidence, authority limits, impact, and validation outcomes.

Also include he-strategy, subagent_policy, capabilities_covered, and capabilities_not_covered.

Procedure

  1. Select exactly one mode unless the user explicitly asks for the full strategy pipeline; ask once only when ambiguity materially changes the artifact.
  2. Resolve he-strategy in ../../references/routing-map.json, build the generic Desktop capability packet, and follow shared subagent policy.
  3. Choose the output path or Do Not Create before writing; do not overwrite artifacts without explicit authority.
  4. Start with 2-3 focused evidence surfaces; widen only when conclusions cannot be proven from the selected set.
  5. Classify .harness artifacts by content shape before path; strategy output is secondary context, not implementation permission.
  6. Apply source-prompt, first-principles contract, XP operating contract, and other shared HE contracts only when their triggers apply.
  7. Apply the context-disposition policy before trimming any section; preserve mandatory markers in-place and move deep detail into routed references.
  8. If evidence is sampled, stale, or narrow, label authority limited and record downstream confidence plus refresh work.
  9. Compress conclusions to decisions that change routing, deletion, investment, anti-drift behavior, or the smallest feedback-producing next slice.
  10. Validate the artifact against the selected mode contract and record each gate as pass, fail, or blocked.

Validation

After skill edits run strict audit, skill_gate.py, OpenClaw, OpenAI format lint, progressive-disclosure lint, family benchmarks, Plugin Eval, and Vale for Markdown prose. For artifacts, verify naming, sources, evidence matrix, confidence, authority limits, and stop/pivot condition. Rerun failed gates. Only mark prose validation blocked if Vale is temporarily unavailable and record a waiver using <!-- vale off -- reason: <reason>; ticket: <ticket>; expires: <date> --> with a corresponding <!-- vale on -->.

Constraints

Redact secrets. Network, destructive filesystem, git write, package install, credential, deployment, and external mutation commands require user approval and active-rule support. Strategy is advisory unless admitted by .harness/linear/**, .harness/reframes/**, .harness/specs/**, or .harness/plan/**.

Execution Boundaries

Generate strategy, review, triage, decision, or core artifacts only. Do not create Linear work, implement recommendations, mutate unrelated pipeline artifacts, or edit runtime/generated projections as canonical source.

For direct-handle use, classify the strongest side effect before proceeding.

Failure Mode

If evidence is missing, mark the conclusion Unknown. If the artifact would create low-value governance, return Do Not Create. If strategy would become implementation by stealth, stop and route to he-reframe, he-linear-plan, he-spec, he-plan, or he-work only after an admitted execution slice exists.

Handoff Rules

  • Hand off to he-spec, he-plan, he-work, he-reframe, or he-linear-plan only when strategy exposes an execution slice.
  • Hand off to humans for ADRs, core invariant changes, strategic deletion, or unresolved instruction conflicts.
  • Use hooks, CI, MCP tools, or validators for enforcement; this skill does not replace those gates.

Accessibility Requirements

Use plain Markdown, short sections, descriptive links, and non-color-only tables.

Gotchas

  • Polished strategy can still be narrow evidence; label coverage limits.
  • Matching old prompt shape is not equivalence; compare covered prompt families, evidence depth, and not-inspected surfaces.
  • Process volume and artifact count are false moat signals until tied to a verified failure or feedback loop.

Anti-Patterns

  • Repeating prior .harness documents instead of compressing them.
  • Creating ADRs or core files for routine implementation details.
  • Producing current-standards claims without current sources or an unavailable evidence note.
  • Letting strategy output authorize implementation.

Examples

  • User asks: "Create a dated JSC-321 repo intent artifact from live source."
  • User asks: "Convert review and triage into evidence-backed strategy."
  • User asks: "Validate whether an ADR is justified, or return Do Not Create."
  • User asks: "Compare what the docs say this repo is for against what the source, command surfaces, tests, validation gates, runtime paths, naming, generated artifacts, and coupling imply."

Output Format

Use the selected mode contract. Include schema_version: 1, source artifacts, facts, interpretations, assumptions, confidence, authority limits, relevant impact, future-agent guidance, validation outcomes, and evidence traceability. Selected outputs must also include the source prompt family status and all mode-specific required sections from ../../references/skills/he-strategy/strategy-output-contract.md, including post-artifact review status where the mode contract requires it.

Stage Arc Boundary

Before artifact writes, mutation, scheduling, handoff, or closure claims, apply ../../references/stage-arc-boundary-contract.md. Structured outputs and handoffs must include stage_arc_boundary with left_arc, active_arc, right_arc, coding_lens, and testing_lens; block when left evidence is stale, active mutation exceeds authority, right-side proof is missing, or a required persona lens is not covered.

References

  • Read when selecting mode output -> ../../references/skills/he-strategy/strategy-output-contract.md
  • Read when architecture lenses or internal canon fallback applies -> ../../references/skills/he-strategy/architecture-lens-canon.md, ../../references/skills/he-strategy/repo-cognition-pipeline.md
  • Read when comparing old prompt workflows -> ../../references/skills/he-strategy/source-prompt-preservation.md
  • Read when validating contract or eval coverage -> references/contract.yaml, references/evals.yaml
  • Read when routing shared HE contracts -> ../../references/deferred-context-index.md
  • Read when resolving helper roles -> ../../references/subagent-call-contract.md, ../../references/subagent-routing.md
  • Read when applying first principles or XP -> ../../references/first-principles-contract.md, ../../references/xp-operating-contract.md
  • Read when applying pragmatic architecture review lenses -> ../../references/pragmatic-programmer-review-contract.md
  • Read when strategy depends on production domain modeling -> ../../references/domain-model-production-contract.md

Do not remove important context for budget trimming; move deep context into reference files and link the relevant route.

Keep looking

Skills are one crate of 328,083. 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.