AI EXTENSION DECISION WORKSHEET

Use this to choose a small change for an observed problem.
This is a worksheet, not an installable skill or access-control configuration.

BLANK WORKSHEET

Task and output:
Observed problem:
Available input and tools:
Standing rule, if needed:
Repeatable method, if needed:
Missing access, if any:
Smallest change:
Expected result:
Failure case:
Observed result after trying the change:
Decision: keep / revise / remove
Next action:

EXAMPLE 1 — FICTIONAL PROPOSAL SCOPE

Task and output: review outputs/proposal-v2.md against the approved brief.
Observed problem: the draft promises a custom booking system.
Available input and tools: agent can read the brief and draft.
Standing rule: check approved scope before making feature promises.
Repeatable method: not required for this single correction.
Missing access: none.
Smallest change: point to the approved scope and add the standing rule.
Expected result: flag the custom-booking promise and cite the scope.
Failure case: do not infer approval when the brief says the decision is pending.
Observed result: not executed; these are illustrative expected results.
Decision: try the smallest change, then record the result.
Next action: review the named draft.

EXAMPLE 2 — FICTIONAL REPEATED REVIEW

Task and output: review recurring proposal drafts.
Observed problem: reviews miss missing pages and unsupported guarantees.
Available input and tools: approved brief, draft and local file access.
Standing rule: drafts remain drafts until the owner accepts a named version.
Repeatable method: compare scope, claims, questions and acceptance criteria.
Missing access: none.
Smallest change: a reusable checklist; package as a skill if repeated use helps.
Expected result: find the missing Contact page and unsupported guarantee.
Failure case: a correct passage should not be marked as an error.
Observed result: not executed.
Decision: test the checklist before choosing a skill package.
Next action: use the deliverable review checklist on one representative draft.

EXAMPLE 3 — FICTIONAL CURRENT ISSUE STATE

Task and output: a weekly note containing the current state of a named issue.
Observed problem: agent cannot read the issue.
Available input and tools: project files, but no authorized source connection.
Standing rule: separate source facts from interpretation.
Repeatable method: only needed if this report becomes recurring.
Missing access: read the named issue through the intended account.
Smallest change: authorized export for a one-off task, or suitable connector/API/MCP.
Expected result: retrieve the right issue and preserve its source reference.
Failure case: missing access is reported, not replaced with a guessed status.
Observed result: not executed.
Decision: verify the read operation before building the report around it.
Next action: check the source and access requirements.

Guide: https://workpulsion.com/guides/skills-vs-mcp-vs-project-instructions/
Do not put passwords, tokens or private client material into a publicly shared worksheet.