Files
scriptorium/docs/roadmap/commandkit-notepad.md

309 lines
17 KiB
Markdown

# CommandKit migration notepad
Status: discussion in progress. This is the working record for agreeing on
Scriptorium's CommandKit migration before preparing a feature roadmap. It
captures direction, open decisions, and discussion notes; implementation scope
and sequencing will follow once the relevant decisions are settled.
## Product context
Scriptorium is a neutral executable interface over PromptKit. Its primary
consumers are applications that need to run stored prompts and consume their
output without importing PromptKit or building an internal prompting workflow.
The CLI and its subprocess contract are therefore central to this migration.
An installation may load many independently maintained collections. The
existing collection model remains the starting point: ConfigKit imports discover
`collection.yml` or `collection.yaml` entry points, each collection declares its
own namespace, and relative imports assemble its profiles and prompts. See the
[configuration contract](../config.md) and
[discovery example](../../examples/config.discovery.yml).
HTTP remains secondary to the CLI discussion. A broader HTTP design review is
planned before v1.0; adopting CommandKit does not itself settle that design.
## Agreed direction
### Full native cutover
Decision recorded 2026-11-19: become a CommandKit-native application, including
`ckconfig` and `ckprompt`, and adopt upstream public conventions and
recommendations as fully as possible.
Backward compatibility with existing CLI conventions is an explicit non-goal.
Legacy command names, flag spellings, parsing rules, report shapes, and exit
semantics do not constrain the new interface. Do not add compatibility wrappers,
legacy aliases, or duplicate inspection implementations solely to preserve the
old CLI. Upstream convenience aliases can be considered on their own merits.
Provide migration guidance in release documentation so consumers can update
their invocations and output handling. Guidance replaces a compatibility layer;
it does not require a transition period with both interfaces. The
[release policy](../policy/release.md#versioning-and-compatibility) governs
release version selection and documentation of breaking changes.
### Ownership and integration
Use CommandKit for command declarations, parsing, help, reporting, and outcomes;
use `ckconfig` and `ckprompt` for their standardized operator interfaces.
Scriptorium should retire the corresponding custom inspection machinery as
those interfaces take over.
PromptKit continues to own preparation and execution semantics. ConfigKit
continues to own configuration resolution. Scriptorium supplies application
configuration and source selection, generation and server commands, process
lifecycle, and application-specific artifact handling. CommandKit adoption
does not introduce workflow or conversation state.
Prefer shared upstream capabilities over new Scriptorium-specific equivalents.
If a required shared capability is missing, discuss an upstream request before
designing a local substitute. Application-specific behavior still belongs in
Scriptorium.
### Initial capability set
Decision recorded 2026-11-19: include configuration comparison, provider smoke
tests, and shell completion in the initial migration, alongside application and
PromptKit configuration scopes, resource inventories and inspection,
requirements discovery, concrete request validation, and prepared-request
preview.
Use the existing upstream capabilities. Configuration comparison should cover
both application and PromptKit scopes, with independent selections for each
side. Provider smoke tests run when explicitly invoked; including them does not
make them startup checks or release prerequisites. Document provider usage and
the distinction between successful diagnostic execution and positive findings.
Shell completion should use upstream command, flag, static-value, and path
completion, with appropriate hints for application-owned inputs. Dynamic
completion of installed resource names is a separate possible enhancement.
Application-specific wiring remains to be designed, but inclusion of these
capabilities is settled. The prepared-request preview decision is recorded below.
### Generation and subprocess requirements
Decision recorded 2026-11-19: optimize the default generation response for the
majority of consumers, who want only the raw resulting artifact. Metadata is
optional and available through the native result reporting choices below.
Prefer native CommandKit reporting and PromptKit result types before introducing
a Scriptorium-specific response format. A failure must produce a nonzero process
status and information explaining the failure. This does not require a new
structured error envelope: native diagnostics are an acceptable way to deliver
that information.
Scriptorium remains a neutral interface to PromptKit. Pass through the errors,
validation outcomes, and result information that PromptKit supplies rather than
adding independent judgments about generated content or redefining upstream
failure semantics. Adapting those outcomes to CommandKit and process exit codes
is application work; deciding what constitutes a PromptKit failure is not.
### Native result reporting and separate report files
Decision recorded 2026-11-19: use a CommandKit `Report` containing PromptKit's
native `RunResult`, with a text template that emits only generated content.
Default/text output serves artifact consumers; standard JSON/YAML formats expose
the native result and its metadata. This composes existing upstream APIs;
ckprompt does not supply a general generation command or prescribe this template.
Include an optional `--report PATH` flag requiring an explicit file destination.
Omitting it creates no separate report file. When requested, save the native
result as JSON through CommandKit's renderer, using the same representation as
inline JSON reporting. The report file's format is JSON independently of the
main result's `--format`; requesting a report does not change the main result's
representation or destination.
Failure to write a requested report is a delivery failure and must produce a
nonzero outcome and diagnostic.
Do not include a configured reporting directory or automatic report collection
in the initial migration. The caller selects the report path and manages its
lifecycle. A configured directory may be considered later if automatic
collection becomes useful; an explicit report path would take precedence.
Native result reports are available when PromptKit returns a result. Operational
failures produce a nonzero exit and native diagnostics. This follows PromptKit's
capabilities and is the intended boundary, not a gap for Scriptorium to address.
Do not synthesize replacement results or add an alternative failure-report
format to compensate for the absence of a native result.
### Output destinations and verbosity
Decision recorded 2026-11-19: use `--output PATH` for the main result's file
destination. Without it, the main result goes to stdout. With it, the selected
representation goes to the file without duplicate stdout output. Use the
spelling `--output`, with no legacy `--out` compatibility alias.
`--format` selects the representation independently of the destination:
default/text emits the raw artifact, while JSON/YAML emits the native result
with metadata. The filename does not select the format. `--report PATH` remains
an additional native JSON report, so callers can save the raw artifact and its
report to separate files.
Adopt CommandKit's upstream verbosity and `--quiet` semantics fully, including
mode selection, message visibility, rendering, and presentation failures. Do
not add Scriptorium-specific exceptions to normal command-output suppression.
Quiet mode does not cancel generation, and failure explanations remain visible
through the native diagnostic channel.
Explicitly requested `--output` and `--report` writes are application-owned file
operations and still occur in quiet mode. Render those files through CommandKit's
writer API; the invocation reporter governs console output. This follows the
upstream distinction between reporting visibility and application effects.
With quiet mode and no requested files, generation still runs while ordinary
result output is suppressed. The [upstream reporting contract][upstream-cli]
owns the shared verbosity rules.
### Prepared-request preview
Decision recorded 2026-11-19: replace `render` entirely with the native
`promptkit show prepared PROMPT_ID` command, enabled through ckprompt's request
preparation integration. Remove `render` without a compatibility alias.
Use the native prepared-run report, standard request flags, diagnostics,
verbosity, and outcomes. Default and explicit text output use the upstream YAML
presentation; JSON remains available through `--format json`. Remove
`defaults.render_format` and any custom prepared-output formatting code left
unused by the migration.
Share application configuration selection, collection loading, and applicable
request construction with generation so previews describe the corresponding
execution request. PromptKit continues to own preparation semantics.
Use the native result stream for preview output; consumers can capture or
redirect stdout to save it. The generation command's `--output` decision does
not itself extend the native inspection commands. File-output convenience does
not justify retaining a separate preview command. Update maintained examples
and release migration guidance to use `promptkit show prepared`.
### Configuration selection and namespaces
Decision recorded 2026-11-19: fully adopt native configuration and namespace
conventions. Expose the `app` and `promptkit` configuration scopes through
ckconfig. `app` inspects Scriptorium's application settings; `promptkit`
inspects the framework configuration assembled from the selected sources.
Collections do not become separate configuration scopes. Application-only
inspection must not require constructing a PromptKit engine.
`--config` consistently selects Scriptorium's application configuration,
including when inspecting the `promptkit` scope. That configuration selects
the framework sources and collection-discovery file. Adopt the standard
`--overlay` spelling and use consistent source-selection and loading rules
across generation, preview, inspection, requirements, and other applicable
commands. Keep the established ConfigKit collection-import model.
Enable the native prompt, profile, and backend namespace selectors on
applicable commands. They independently scope short caller-supplied resource
IDs; qualified IDs remain exact selections. Namespace selection does not
rewrite stored references or change collection discovery. Keep configuration
scope, resource namespace, and profile selection distinct.
Use native inventory filters, including namespace selection and
`--all-namespaces`, and retain upstream default listing behavior. Support
standard temporary configuration overrides where the upstream loader supports
them. The reviewed PromptKit configuration scope does not support those
overrides; do not add local override semantics to compensate.
Do not add configurable default namespaces in the initial migration. Use
qualified IDs and explicit namespace flags when selecting namespaced resources,
while retaining upstream behavior for unqualified IDs. Configurable defaults
can be considered later if a need arises.
### Native inspection reports and consumer requirements
Decision recorded 2026-11-19: adopt native inspection reports and the native
`promptkit requirements show` interface for Narratio's early dependency
discovery. Its consumer needs and acceptance scenarios are recorded in the
[aligned Narratio proposal](narratio-feature-request.md). This direction is
accepted; Scriptorium integration remains to be implemented.
Consume native report fields without a Scriptorium-specific schema-version
envelope. Use native nonzero outcomes and diagnostics for invocation failures;
JSON report selection does not require JSON errors. Neither a separately
versioned envelope nor programmatic invocation-failure categories blocks this
migration. A future concrete need for either belongs in an upstream discussion.
PromptKit owns requirement classification, discovery completeness, and structured
discovery issues. Completeness is distinct from satisfied bindings or execution
readiness; Narratio owns comparison against its planned inputs. Keep early
discovery separate from concrete request validation. Use current exact resource
identities and native input/variable metadata, without restoring obsolete prompt
versions or inventing input schemas or variable-default semantics.
Standard prompt inspection includes captured messages; prepared preview includes
rendered content. Adopt these native reports without rebuilding Scriptorium's
previous summary projections, and explain the changed presentation in migration
guidance. Requirements discovery retains its distinct upstream content boundary.
## Upstream baseline
The initial review covered CommandKit **v0.7.0**, ckconfig **v0.9.0**, and
ckprompt **v0.10.0**. These are a review baseline, not a final dependency choice
for implementation. The [upstream consumer guide][upstream-guide] describes
their composition and supported extension points.
| Component | Relevant capabilities |
| --- | --- |
| CommandKit | Typed flags and operands with explicit-value presence, generated help and version, shared reporting and outcomes, optional shell completion. |
| ckconfig | Configuration scopes; validation, effective values, sources, explanations, and comparisons; temporary overrides where supported by the loader. |
| ckprompt | Prompt, profile, backend, namespace, and collection inventories; resource inspection; requirements discovery; optional concrete request validation, prepared-request previews, and provider smoke tests. Reusable request flags also serve application-owned commands. |
Scriptorium currently uses PromptKit v0.23.3 and ConfigKit v1.2.1. The reviewed
ckprompt release declares PromptKit v0.23.2 and ConfigKit v1.2.1. Validate the
chosen dependency combination while retaining Scriptorium's adopted PromptKit
fixes; the older declared minimum is not a reason to downgrade.
The [upstream CLI reference][upstream-cli] owns the detailed conventions. The
[current Scriptorium CLI](../cli.md) and
[subprocess contract](../integrations/subprocess.md) describe the starting point,
not requirements to reproduce during migration.
## Open decisions
### 1 Generation integration details
The consumer requirements, native result reporting, output destinations, and
verbosity decisions above are settled. The remaining work is to specify their
integration in the feature roadmap.
Map PromptKit's typed errors and reported outcomes into CommandKit conventions,
preserving its distinction between a completed content check reporting
`ValidationFailed` and an operational error. Select any necessary domain exit
code without reusing the current validation-failure exit 2, which conflicts with
CommandKit's usage convention. Also specify artifact/report delivery failures
using CommandKit's presentation-failure behavior where applicable. These are
adapter mappings, not a new application validation policy.
Signal cancellation and the whole-operation deadline remain application
responsibilities. Decide their mapping to the CommandKit invocation context
and PromptKit operation timeout, separately from the standard per-attempt
timeout flag.
### 2 Application commands and standard request flags
Decide the final generation command operands and application-specific flags
around the reusable ckprompt request flag groups. Use their standard names,
value-presence behavior, and parsing semantics wherever applicable.
Migration guidance will need to explain changes such as separate literal-text
and file-input flags, endpoint and timeout naming, and repeated mapping values.
In particular, upstream's optional `--input` alias denotes literal text, while
Scriptorium's existing flag denotes a file. Choose any upstream convenience
aliases deliberately and document the new meanings; preserving old meanings
is not a design requirement.
## Continuing the discussion
Record discussion notes under the relevant open decision. When a choice is
settled, move its conclusion and rationale into the agreed direction with its
decision date, and remove superseded alternatives. Add new questions when a
concrete product need emerges.
Once the command surface, consumer outputs and outcomes, loader and namespace
behavior, and any upstream gaps are settled, fold the decisions into a feature
roadmap. That roadmap will define implementation sequence, removal of replaced
CLI code, appropriate validation, and migration documentation. This notepad does
not prescribe implementation phases or treat the open proposals as commitments.
[upstream-guide]: https://forge.maximumdirect.net/base-layer/commandkit/src/commit/d330d7637f2e7507eca931a899397222c82bc892/docs/consumers/guide.md
[upstream-cli]: https://forge.maximumdirect.net/base-layer/commandkit/src/commit/d330d7637f2e7507eca931a899397222c82bc892/docs/cli.md