Skills lifecycle reference
Improve one workflow through human + AI Agent work, then repeat with evidence. This is the practical experiment supporting the 6 Months Roadmap, not a mandatory skills program or permission to deploy the long-term autonomous Target. Use context, coaching, test/review feedback or deterministic automation if they better address the bottleneck.
Read the arrows as handoffs, not a checklist of page approvals. Governance bounds every activity. Measurement supplies selected definitions throughout. Quality establishes candidate fitness; Adoption tests usefulness in real work; Maintenance returns failures and changed conditions to qualification.
The 6 Months Roadmap owns EXP-01 and the month-by-month plan; this reference only explains reusable lifecycle handoffs within the broad Roadmap.
Delivery first: one change, from intent to accepted work
Illustrative Delivery workflow throughout these pages — not an Eldorado pilot or measured result: a Product Engineering developer uses an AI Agent to prepare a bounded code change and regression tests. An independent reviewer checks the exact change and evidence. Existing release authority remains unchanged; the skill cannot approve or deploy its own output.
First trace codebase structure and existing principles → system-logic source → tests → builds and packaging → deployments. Identify the friction to improve—for example, incomplete regression evidence that sends changes back for rework. Verify that this is the actual problem before choosing a skill.
Adoption can extend to other SDLC phases; their order is not yet decided. Delivery keeps improving as adoption expands.
Carry evidence forward, not assumptions
- Understand and baseline. Product Engineering maps the selected workflow and desired improvement. Select relevant KPIs from Measurement; Baseline records current outcomes, acceptance rules and explicit gaps. A baseline describes work; it grants no approval.
- Choose the smallest candidate. If repeated agent guidance is the bottleneck, use Governance to connect repository understanding → change/test preparation → independent verification. Specify triggers, inputs, outputs, handoffs, pinned dependencies and forbidden actions. This skills topology is more than a catalogue; a deterministic fix may need no skill.
- Qualify before exposing real work. Quality compares the candidate with declared current practice. If no approved skill exists, approved-revision regression is N/A with reason; test removal and recovery to current practice. Once an approved revision exists, add that comparator. The independent Quality Engineering reviewer function accepts evaluation evidence; the owning team's designated approver authorizes only the tested pilot envelope, with security review where risk requires it. Failed checks or missing authority mean rework or hold, not an automatic pilot.
- Pilot and compare. Adoption applies that revision to a bounded real-work cohort, retaining the manual path. Compare accepted outcomes, quality, review/rework cost and feedback with the starting position. Live tasks and qualification trials remain separate populations.
- Decide and maintain. The owning team's designated workflow owner decides retain, improve, stop or propose expansion from the comparison and independent evidence. Maintenance keeps retained revisions supported and feeds defects into requalification. A new scope returns to Baseline; a changed revision returns to Quality. The new current situation starts the next loop.
These are recommended functions, not evidence of actual departments or appointed people. The architect may build and coordinate; a willing developer/reviewer and actual workflow owner agree contributions. One worksheet plus existing PR/CI records can suffice. Before work starts, record actual people, acceptance rules and permitted effects. Missing assignments keep the affected decision on hold.
Where each topic helps—and what it gives the next activity
| Topic | Use it when | Carry forward |
|---|---|---|
| Measurement | Choosing or interpreting evidence at any point | Relevant contracts for Workflow effectiveness and Skills effectiveness & health; no mandatory full catalogue |
| Baseline | Understanding a workflow or a material scope change | Workflow map, selected definitions, evidenced starting position or gaps |
| Governance | Designing, approving, distributing or withdrawing any revision | Named accountability, authority envelope and exact-revision decision record |
| Quality | Introducing or changing skills, clients, models or dependencies | Versioned evaluation results, failures, limits and independent recommendation |
| Adoption | Introducing a qualified candidate into bounded work | Comparable outcome report, user feedback and retain/improve/expand/stop decision |
| Maintenance | Keeping a supported workflow dependable | Reviewed revision, regression cases, recovery evidence or retirement |
The six topics are connected lifecycle activities, not six rigid sequential approval gates. The sidebar is a reading map: you can consult Measurement before, during or after any activity.
Proven skills and working practices may support Beyond 6 Months. Stage 01 can remain valuable indefinitely; progression requires a separate case for authority and readiness.
Start here: Understand and baseline one workflow →. Consult the Measurement registry when selecting evidence.