| name | compound-learning-capture |
| description | Use when a verified fix, feature, audit, incident, release, or repeated workflow produces reusable learning that should improve future APIVR plans, reviews, reports, or release decisions without becoming a parallel source of truth. |
| activation | Activate when the description trigger applies to the current task. |
| required_inputs | Task request, relevant repository context, constraints, and authority dependencies. |
| required_outputs | Skill-specific artifact, verification evidence, canonical verdict, and next action. |
| authority_dependencies | 00_start_here/SOURCE_OF_TRUTH.md; 10_governance/APIVR_EXECUTION_LIFECYCLE.md; 10_governance/source_of_truth/Elite_Build_Goals_v3.md. |
| evidence_requirements | Executed checks or an honest Unknown, Not Run, or Blocked state for every material claim. |
Compound Learning Capture
Use this skill after evidence exists. It turns a solved problem into a small reusable learning artifact that future agents can find, verify, and refresh.
Do not capture guesses, preferences, or unverified lessons as durable knowledge. A compound learning entry must point to evidence, name its scope, and remain subordinate to APIVR, the Elite Build Goals, and canonical kit files.
When To Use
Use for Standard and above work when one or more are true:
- A bug, incident, security issue, release problem, or integration failure was solved.
- A plan or review found a reusable decision pattern.
- A repeated workflow produced a better stop condition, evidence rule, or routing rule.
- A provider, API, hosting, automation, UI, reporting, security, or build system behavior was learned.
- A completion report contains a lesson that would prevent future rework.
Do not use for one-off trivia, unverified hunches, project secrets, credentials, private customer data, or content that belongs only in an issue tracker.
APIVR Placement
Compound learning is normally an APIVR Phase 6 Re-Audit output.
- Audit: identify the observed problem or decision.
- Plan: decide whether the lesson is reusable and where it belongs.
- Implement: create or update the learning artifact.
- Audit Implementation: check for duplication, overgeneralization, and stale-source risk.
- Verify Implementation: confirm links, evidence, scope, and routing.
- Re-Audit: decide whether to keep, consolidate, or schedule refresh.
Decision Flow
flowchart TD
A["APIVR work produced evidence"] --> B{"Reusable beyond this task?"}
B -- "No" --> C["Record only in completion report"]
B -- "Yes" --> D{"Evidence state is Verified or accepted Conditional PASS?"}
D -- "No" --> E["Keep as Unknown or Not Run in evidence ledger, not durable learning"]
D -- "Yes" --> F{"Would this create duplicate source of truth?"}
F -- "Yes" --> G["Update canonical skill, knowledge, template, or audit file instead"]
F -- "No" --> H["Create solved problem learning entry"]
G --> I["Add provenance and refresh trigger"]
H --> I
I --> J["Route through knowledge-refresh-and-drift-control later"]
Required Fields
Use 60_templates/SOLVED_PROBLEM_LEARNING_TEMPLATE.md.
- Problem pattern.
- Context where it applies.
- Evidence that proved the lesson.
- Decision or action that worked.
- What failed or was rejected.
- Canonical files affected or referenced.
- Future trigger for using the lesson.
Provider webhooks are flaky. Add retries next time.
When provider webhooks can be delivered more than once, require an idempotency key test before release. Evidence: duplicate-delivery test passed, sandbox replay produced one payment record, and logs showed no secret exposure. Applies to payment webhooks and other write-producing callbacks. Canonical routing: `skills/external-api-integration/SKILL.md`, `skills/test-driven-development/SKILL.md`, and `10_governance/RELEASE_GATES.md`.
