Technical format: Skill
Guided Practice
Practice one work judgment with bounded help, then return control to the person.
What Guided Practice does
For a person who explicitly wants to learn while working, elicit a small attempt and reason, give one diagnostic question or bounded contrast, and reduce help only when the actual attempt supports it. An unaided transfer attempt remains a separate observation.
- For
- People and teammates who explicitly choose practice while reviewing a work artifact.
- Bring
- An explicit practice request, one authorized artifact, an observable target, scope and round/time budgets.
- Action
- Starts with the person’s attempt, offers bounded feedback, checks their reasoning and returns a keep/revert/leave-unchanged decision.
- Output
- An actual attempt and reason, bounded feedback and a human decision; optional unaided practice is recorded separately.
Runtime and prerequisites
A portable method in your existing Agent Skills host. An optional Python 3.11+ helper checks declared stop facts from two records; it is not a coach runtime.
- Use an existing skill-capable assistant with authorized access to the chosen artifact. No other capability, new package or provider configuration is required.
- Python 3.11+ is needed only for the optional read-only helper. Its file CLI uses POSIX no-follow reads; the dictionary API is independent.
Works on its own: Practice is optional and self-contained. Direct execution remains the default when practice was not requested; existing coaching modes and budgets remain authoritative.
Natural stop: Stop at user stop or “just finish”, target, agreed cap, stalled gain, missing input/authority or a failed/unknown critical gate. Return the current artifact for human decision. No standing memory, transcript or reminders are collected; notes require separate opt-in.
Example output
A fictional practice exchange with an initial attempt, one diagnostic question and an explicit human decision.
Fictional illustration, not a result from your task or measured user evidence.
Limitations
Declared stop records do not prove consent, authority, actual attempts, learning or correct counters. The helper reads no artifact and launches no task. Human learning and outside-user usefulness have not been measured.
Explore by kind of work
Another part of the work.
Back to the collection
Related tools & research
Sources to compare or explore alongside this capability. A listing does not imply an integration, partnership or endorsement.
Open source 1 source
Anki
Spaced flashcard recall serves a different job from practicing a judgment in an existing work artifact. There is no Anki integration or scheduling here.
Difference: No code, content or assets copied. Anki is AGPL 3-or-later with component exceptions; see its LICENSE.
Reviewed
Research & guidance 2 sources
Roediger and Karpicke (2006), Test-enhanced learning
The author abstract contrasts immediate and delayed prose recall. It motivates measuring unaided transfer separately; it does not validate this method or work-judgment coaching.
Difference: Abstract/bibliography reviewed; no full-text access or efficacy extrapolation. DOI 10.1111/j.1467-9280.2006.01693.x.
Reviewed
Buçinca, Malaya and Gajos (2021), To Trust or to Think
An AI-assisted decision experiment reports a tradeoff between reduced overreliance and subjective burden. Opt-in, direct-execution and stop choices must remain available here.
Difference: Author abstract reviewed; no implementation copied or benefit claimed for this method.
Reviewed
Commercial tools 1 source
Khanmigo
A hosted educational tutor and teaching assistant. Guided Practice is a caller-applied work-practice procedure, without a hosted classroom service or curriculum.
Difference: Research only; no account, integration, copied prompts or proprietary content.
Reviewed