mirror of
https://github.com/magnus919/agent-skills.git
synced 2026-09-18 15:06:28 +03:00
Magnus HedemarkandGitHub
cff17c5974
* fix: SkillOpt Epoch 1 — forward-deployed-engineering optimization Inline the nine-stage contract table into SKILL.md (required question, minimum output, stop condition per stage) with template links and the entry-evidence rule; dedup the stage table out of references/lifecycle-and-artifacts.md into a pointer. Name agent-evals-and-observability and production-readiness inline in the applied-AI release gate (loading protocol step 5). Validated: 2/2 held-out edits accepted (non-regression, all-pass baseline), repo validators green (validate-skills, validate-skill-quality, validate-bundles, validate-evals). * fix: SkillOpt Epoch 2 — forward-deployed-engineering optimization Add a 'Where to enter the lifecycle' table (existing state -> entry stage, with the neckbeard route for bounded changes) and the entry-evidence rule for mid-stream joins. Replace the flat 'When not to use' list with a proactive Scenario | Reach for | Why routing table covering the six boundary routes. Validated: 2/2 held-out edits accepted (non-regression, all-pass baseline) plus a regression probe on epistemic labels; repo validators green. * fix: SkillOpt Epoch 3 — forward-deployed-engineering optimization Add references/worked-example-engagement.md, a fully synthetic depth calibration artifact showing the charter, evidence-labeled ledger, stage handoff, evaluation and release decision, adoption scorecard, outcome measurement record, and productization record for one engagement. Add a File map row, enumerate the templates row (surfacing engagement-status), and add a depth-calibration pointer in the Lifecycle section. Validated: 2/2 held-out edits accepted (non-regression, all-pass baseline); repo validators green; sanitization scan clean (no private identifiers).
forward-deployed-engineering
Carry an embedded technical engagement from an ambiguous need to an adopted, measurable capability and a deliberate generalization decision.
Why Install This Skill
This bundle is designed to guard against continuity risks between disciplines: discovery that leaves builders without usable context, prototypes mistaken for production systems, deployments declared complete without adoption, and field patterns transferred without evidence or an accountable receiving owner. It gives an accountable lead a durable contract across those boundaries.
After installation, an agent can frame the engagement and decision rights, route specialist work without duplicating it, preserve assumptions and risks, diagnose adoption and outcome gaps, and decide whether local work should remain configuration, become reusable, transfer to another owner, or be retired.
What You Get
| Path | What it provides |
|---|---|
SKILL.md |
Lifecycle trigger, boundary, loading protocol, stop rules, and file map |
manifest.yaml |
Machine-readable stages, routed skills, outputs, handoffs, and conflicts |
references/discovery-brief.md |
Overlap audit, source decisions, and design-risk rationale |
references/source-index.md |
Recoverable primary sources, supported claims, limitations, and refresh rules |
references/lifecycle-and-artifacts.md |
Stage contracts and continuity artifacts |
references/route-selection.md |
Per-stage specialist entry conditions and direct-routing boundaries |
references/authority-and-escalation.md |
Decision rights, constrained-environment discovery, and escalation rules |
references/adoption-and-measurement.md |
Adoption diagnosis, outcome evidence, and applied-AI release evidence |
references/generalization-and-productization.md |
Classification and productization decision model |
references/communication.md |
Evidence-labeled status, handoff, and escalation patterns |
templates/ |
Ten templates matching the declared continuity, decision, handoff, measurement, and learning outputs |
evals/evals.json |
Fifteen output-quality cases covering lifecycle, authority, and adjacent-skill boundaries |
Quick Start
- Load
SKILL.mdand createtemplates/engagement-charter.md. - Record the stakeholder workflow and unknowns before proposing architecture.
- Treat
manifest.yamlstage skills as candidates and applyreferences/route-selection.mdbefore loading one primary specialist. - Carry the ledger through deployment, adoption, measurement, and generalization.
Triggers
- An embedded technical lead must own discovery through production adoption.
- A stakeholder request is ambiguous and implementation context must be discovered.
- A prototype must become a verified, deployed, adopted capability.
- An applied-AI engagement needs eval-driven release and field feedback.
- A local configuration or workflow may warrant a reusable pattern or product capability.
- Adoption, workflow impact, and field learning must remain connected to delivery.
Do not trigger for a bounded repository bug, standalone specialist task, product portfolio governance, ongoing SRE ownership, platform operation, or advisory work that ends before implementation and adoption.
Requirements
- No API keys, services, or network dependencies.
- Compatible with Agent Skills harnesses that support file reading, writing, and skill loading.
- Routed catalog skills must be installed, including discovery, evaluation, production readiness, deployment, adoption, measurement, security, privacy, and constrained-environment specialists.