# Shared debug binding implementation plan

Status: in progress; Stages 1–4 complete.
Baseline: `6727d22ce6e1ac668b56a3e6d8fa5f7a06e54d88`.
Execution audience: Luna coding agent.

## Target and execution instructions

The [feature roadmap](shared-debug-binding.md) owns the accepted behavior,
intent, examples, and scope. OD-1 and OD-2 are resolved. The original
Audita request is motivation, not an additional
capture policy for CommandKit. This plan owns implementation order and checks.

Implement Stage 1 through Stage 7 in order as one end-to-end sprint after the
user approves implementation. Stages are work units, not approval checkpoints
or separate agent prompts. Continue through internal dependencies without
requesting permission after each stage. Intermediate stages need not compile
or leave a coherent repository. Record checks that depend on later stages as
pending, then resolve them before finishing; do not add temporary compatibility
code just to make an intermediate stage pass. Investigate and fix failures
within the selected scope. Surface a genuine contradiction rather than silently
changing the target.

Start with [development guidance](../development.md) and follow the
[agent](../policy/agents.md), [architecture](../policy/architecture.md),
[testing](../policy/testing.md), and [documentation](../policy/documentation.md)
policies. Use graph discovery for code as directed by AGENTS.md. Preserve
unrelated work and the ignored workspace files. Record actual completion and
verification here as work lands. Commit completed nonempty units with plain
English messages; dependent stages may share a commit. This sprint does not
publish releases, push refs, or implement Audita artifact capture.

## Implementation decisions

- Accept the existing `BoolFlag` form named `debug`, including ordinary captured
  and renamed boolean bindings. Its private `boolean` property identifies the
  constructor semantics; a matching name alone does not. Retain validity,
  duplicate-name, short-spelling, and wiring checks. Other reserved reporting
  bindings remain rejected. There is no new public constructor or subscription.
- Retain every consumer entry in `commandmeta.Command.Flags` at its original
  index, including shared debug. Retain the core reporting declarations in
  `CoreFlags`. Produce one effective flag list for backend registration,
  token interpretation, and completion, combining only the selected debug
  overlap. Never remove a consumer entry to deduplicate parser registration.
- Build that effective list through one small helper in
  [internal/commandmeta](../../internal/commandmeta/metadata.go), used by the
  backend and completion. Copy metadata rather than changing captured
  declarations. Combine core debug usage with application usage, short spelling,
  requiredness, and completion settings directly. Keep the core reporting
  default false; preserve the application default in its original binding.
  When the application default is true, label it separately in help and explain
  that reporting still defaults to ordinary. Exact prose and helper names are
  implementation details; no generic metadata-merging facility is selected.
- When a debug binding participates, honor its completion hint just as for an
  ordinary application boolean, including no completion or an empty choice list.
  With no application binding, keep core's existing boolean suggestions. Hints
  never constrain parsing. A command-local short spelling belongs only to that
  command; existing flag aliases retain their existing scope and collision rules.
- Feed the single parsed occurrence sequence to reporting reduction and its
  indexed raw debug values to ordinary typed preparation. Defaults and
  requiredness are application input rules, not reporter selection inputs.
- Define public `ReportingMode uint8` and the four selected constants with
  ordinary as zero. Use that type for the existing reporter mode field, replacing
  the private mode type/constants rather than adding a second mutable selection.
  Keep the internal parser's neutral type and translate at `applySelection`.
  `Invocation.ReportingMode()` reads reporter state under its existing mutex,
  returns ordinary for a zero invocation, and remains readable after finalization.
  No public type aliases an internal/backend type; no setter or second public
  accessor is added. Keep the existing reporting callback reentrancy contract.

## Stage 1 — Expose the resolved reporting mode

**Complete.** The invocation exposes the existing reporter selection, including
zero and finalized values. Focused reporting and public accessor checks passed
with Go 1.27.0 and offline module resolution; concurrent reads also passed with
`-race -count=1`.

Primary files: [reporter.go](../../reporter.go),
[identity.go](../../identity.go), [execute.go](../../execute.go).

1. Define `ReportingModeOrdinary`, `ReportingModeQuiet`, `ReportingModeVerbose`,
   and `ReportingModeDebug` on the public type. Replace the private mode
   representation in reporter state, visibility switches, and selection mapping.
   Keep ordinary as the zero value. Update tests referring to removed private
   identifiers without changing their behavioral expectations.
2. Add `func (inv Invocation) ReportingMode() ReportingMode`. Handle a missing
   reporter/state as ordinary; otherwise read the existing field under its
   mutex. Finalization must preserve the mode while releasing its current writer
   references. Each execution continues to create its own reporter.
3. Preserve execution order: reduce explicit occurrences and call
   `applySelection` before typed input preparation and the handler. Do not
   infer mode from bound values, output, format, or progress.
4. Document the method's zero value, stable handler/finalized selection, and
   compatibility with concurrent reporting. It reads selection, not delivery
   success; it adds no concurrent App dispatch or writer-lifetime guarantee.
5. Use `shared_debug_external_test.go` (package `commandkit_test`) as the public
   behavior owner for this API and the shared-binding table in Stage 3. Add
   checks for all four modes, a zero invocation, retained invocation after
   return, and a subsequent execution with no flags.
   Exercise concurrent mode reads alongside reporter operations using bounded
   goroutines and explicit synchronization, not sleeps or concurrent dispatch.
   Reuse an existing safe in-memory destination and existing visibility checks.

Complete when the accessor observes the existing selection without side effects,
and concurrent use and retained invocations preserve the selected result.

## Stage 2 — Recognize shared boolean bindings and compose metadata

**Complete.** Declaration validation accepts only boolean debug bindings and
retains other reservation and wiring checks. Effective metadata combines the
shared option without removing consumer entries. Focused validation and metadata
tests passed offline with Go 1.27.0; backend adoption follows in Stage 3.

Primary files: [command.go](../../command.go), [binding.go](../../binding.go),
[parser.go](../../parser.go), and
[neutral metadata](../../internal/commandmeta/metadata.go).

1. In declaration validation, allow the reporting reservation exception only
   for a valid boolean binding named `debug`. The existing boolean property is
   carried by `BoolFlag`, `CaptureFlags`, and `RenameFlag`; it is not set by
   `ValueFlag` or `OccurrenceFlag`. Preserve all ordinary validation, including
   missing setters, duplicate debug bindings, other reserved names, duplicate
   shorts and reserved `-h`. Do not accept a custom converter because it returns
   a Go bool.
2. Keep the bridge's consumer metadata and indexes intact. Add the effective
   list helper described above: retain ordinary consumer flags, replace the
   participating debug entry with the combined declaration, then append the
   remaining core flags. Only the boolean debug overlap is combined. The helper
   does not turn invalid duplicate application declarations into valid ones.
3. The combined declaration remains boolean with core's false reporting
   default. Use application short spelling, requiredness, and completion hint;
   combine usage text and label a differing application default. The original
   consumer declaration keeps its own default for typed preparation. Backend
   raw input remains empty on omission; metadata must not synthesize an explicit
   occurrence or presence.
4. Adjust the reservation tests in [flag_aliases_test.go](../../flag_aliases_test.go)
   to distinguish accepted `BoolFlag` from still-rejected string/custom-parser
   declarations. Add focused checks for successful registration and duplicate
   or invalid wiring rejection, retaining existing transactional registration
   assertions. Test the effective metadata at its package boundary where that
   gives distinct protection; do not snapshot private slice layouts wholesale.

Complete when declarations express the agreed exception and one effective debug
option can coexist with unchanged consumer indexing and application defaults.
Backend-dependent checks may remain pending until Stage 3.

## Stage 3 — Parse one debug option into both consumers

**Complete.** Backend registration and interpretation use effective flags while
consumer values retain their original indexes. Public mode/value cases, aliases,
invalid values, option-looking values and terminated operands pass, as do focused
backend parsing checks, offline with Go 1.27.0.

Primary files: [backend parse.go](../../internal/cobra/parse.go),
[backend flags.go](../../internal/cobra/flags.go),
[parser.go](../../parser.go), [binding.go](../../binding.go).

1. Use the effective list in `buildTree` registration and `Session.Parse` token
   interpretation. Iterate it once when queuing entered spellings; do not queue
   the same raw debug value twice. Retain existing pflag boolean no-value syntax,
   raw-value collection, alias interpretation, and failure precedence.
2. Extract `FlagValues` and `FlagSpellings` by iterating the original consumer
   `Flags`, looking up each effective backend flag by canonical name. This
   delivers shared debug to its original index while core-only flags still
   occupy no consumer index. Preserve the occurrence log's canonical names,
   entered spellings, and order, including interleaving with other flags.
3. Keep `reduceReportingOccurrences` authoritative for visibility and invalid
   reporting values. Use ordinary `BoolFlag.prepare` for the application value.
   Do not add a setter path in the reporter, rescan argv, or reconstruct mode
   from the final boolean. Reporting must still work on unbound commands,
   root/group help, version, and discovery routes.
4. At the backend boundary, extend
   [parse_test.go](../../internal/cobra/parse_test.go) and existing flag grammar
   coverage only where needed to establish shared extraction, stable indexes
   with ordinary flags before/after debug, and one occurrence stream through
   aliases/short spellings. Do not duplicate the facade behavior table here.
5. At the public command boundary, extend `shared_debug_external_test.go` with
   the roadmap's mode/value/presence table, observing both typed input and the
   new accessor. Check invalid/empty boolean input prevents factories and handler
   execution as appropriate under existing preparation timing. Keep tokens consumed as
   another flag's value and operands following `--` out of reporting reduction.

Complete when debug feeds normal typed preparation and mode selection from one
parse, without changing unrelated input indexes, grammar, or failure handling.

## Stage 4 — Preserve composition, defaults, and alias behavior

**Complete.** Existing capture and rename machinery retains the boolean
identity and requiredness without production changes. Focused public tests
cover true defaults, explicit false, direct and captured requiredness, both
application methods, renaming in both directions, and an alias to the local
short spelling; capture and alias tests pass offline with Go 1.27.0.

Primary files: [flag_values.go](../../flag_values.go),
[binding.go](../../binding.go), [flag_aliases.go](../../flag_aliases.go).

1. Review the shared binding through `CaptureFlags`, `FlagValues.ApplyTo`, and
   `ApplySuppliedTo`. Retain original binding identity, independently selectable
   destinations, and existing conversion timing. Capture already preserves the
   boolean property; avoid adding another store or observer if no change is needed.
2. Preserve `deferredRequired`: direct bindings enforce required input during
   ordinary preparation, while captured bindings check requiredness when selected
   for application, as they do today. An explicit false satisfies presence.
   A true application default with `Set=false` must not satisfy requiredness or
   turn on reporting. `ApplySuppliedTo` must keep its ordinary omission behavior.
3. Keep shared behavior determined by the current binding name and boolean
   semantics. Renaming debug to an ordinary name yields an ordinary independent
   boolean and stops observing core debug; renaming an ordinary boolean to debug
   uses the same accepted attachment. Do not leave a sticky sharing marker.
4. Review canonical short spellings and alias compilation together. A local
   debug short and aliases targeting its canonical debug input must feed the
   same occurrence stream. Preserve existing collision errors, scope, and
   registration rollback; do not add short spellings to unrelated commands.
5. Extend [flag_values_test.go](../../flag_values_test.go), existing rename and
   alias tests for representative captured application, true default/explicit
   false, requiredness, renamed-away behavior, and an alias targeting the local
   short spelling. Use the public table from Stage 3 for visibility behavior
   instead of repeating it in each composition test. Modify production code
   here only where the selected behavior actually needs a change.

Complete when shared binding behaves like an ordinary reusable boolean across
capture, selection, application and rename, with visibility still independent.

## Stage 5 — Present one debug option in help and completion

**Complete.** Completion uses the same effective declarations as parsing and
help. Canonical and alias help and usage failures expose the combined option;
focused tests cover local shorts, requiredness, application defaults, choices,
file and directory hints, boolean validation, and idle application callbacks.

Primary files: [completion_flags.go](../../completion_flags.go),
[internal/cobra/parse.go](../../internal/cobra/parse.go),
[internal/commandmeta](../../internal/commandmeta/metadata.go).

1. Make `completionFlags` use the same effective declarations as the backend,
   alongside the existing help flag. Eliminate the duplicate debug long option
   without suppressing the application's local short spelling. Keep canonical
   suggestion and existing alias interpretation policies; this is not alias
   discovery work.
2. Ensure canonical command help, explicit help targets, alias help, and usage
   failures show the combined debug option once. Core usage and application
   usage must both be intelligible; a true application default is labeled as
   application input, alongside the ordinary reporting default. Retain existing
   required markers and help's exemption from required-input enforcement.
3. Preserve application completion hints, including choices, no completion, and
   file/directory hints under ordinary handling. An arbitrary suggested value
   remains subject to boolean parsing. Keep normal unbound core suggestions;
   do not introduce a debug-specific completion rejection rule.
4. Use existing completion query tests and help tests to check one long debug
   suggestion/row, the local short, a true default/required help case, explicit
   boolean value completion, and one custom hint. Assert meaningful fragments
   and counts for the effective option, not a complete help snapshot or exact
   new explanatory wording. Queries and help must not call input factories,
   setters, `CheckInput`, handlers, or providers. Retain help's existing conversion
   of supplied input for syntax validation; completion continues to use metadata.

Complete when help and completion describe the same single parsed option and
clearly distinguish application defaults from reporting selection.

## Stage 6 — Document the implemented contracts

1. Update `BoolFlag`, `FlagMeta`, binding composition, and Invocation GoDoc with
   the supported debug exception and mode API. Explain defaults, presence,
   requiredness, short spellings, and retained mode without making all reporting
   names bindable or treating mode as delivered-output evidence.
2. Update the [CLI reference](../cli.md#placement-and-aliases) where it describes
   reserved names and command help. Keep API wiring in the
   [consumer guide](../consumers/guide.md#invocation-reporting). Add a small
   compiling example showing `BoolFlag` on `debug` and the mode query, using
   `--debug --quiet` to explain why their results differ. Applications still own
   artifact creation and other effects.
3. Add the feature and consumer impact to the next core minor candidate note,
   `docs/releases/v0.6.0.md`, using the release and OSS policies for its title
   and immutable licensing footer. Core v0.5.0 is published and must not be
   rewritten. If version state has changed before implementation, select the
   next unused minor through the release procedure; this is release bookkeeping,
   not a new decision about this feature. No tag or publication is part of this
   sprint. Integration dependencies and upstream versions need no upgrade here.
4. Update [internal guidance](../internal/overview.md) only where the effective
   flag view and authoritative reporter mode change its explanation. Validate
   changed links, fragments, Go snippets, and prose against the final code.
   Do not introduce a new ADR for this addition.

Complete when GoDoc owns the precise API, operator guidance explains the flag,
and consumer guidance distinguishes input policy from reporting selection.

## Stage 7 — Verify the assembled sprint and record completion

1. Review every target/checklist item in the feature roadmap against the final
   implementation and evidence. Ensure tests that formerly rejected all reserved
   reporting bindings now allow only the selected boolean form. Review capture
   and requiredness separately from the direct binding cases.
2. Run focused tests as the stages become testable, then the independent core
   suite and the aligned three-module suite with Go 1.27.0 and cached dependencies:

   ```sh
   GOWORK=off GOTOOLCHAIN=local GOPROXY=off go test -mod=readonly ./...
   GOTOOLCHAIN=local GOPROXY=off go test -mod=readonly ./... ./integrations/ckconfig/... ./integrations/ckprompt/...
   ```

   Use the ignored workspace described in the developer guide. The aligned
   integration run exercises this changed core; independent integrations still
   select released core v0.5.0 and cannot establish that boundary by themselves.
   Do not upgrade manifests or commit local replacements to run these checks.
3. Run the new focused concurrent mode-read check under `-race -count=1` when
   supported; choose its actual test name after implementation. No full race
   matrix, timing assertion, live provider, or Audita processing run is needed.
   Record unavailable optional tooling accurately and continue with supported
   verification; never describe an unavailable check as passed.
4. Run gofmt on changed Go files; inspect diffs and verify current Markdown links
   and fragments, snippet compilation, and tracked/staged prose whitespace using
   the developer guide. Use a temporary `GOCACHE` if the default is unwritable.
   Reuse passed checks for unchanged content; repeat for actual changes or failures.
5. Record completed stages, actual checks, limitations, and residual release
   preparation here. Mark the feature checklist complete only with its evidence.
   Retire completed planning records through the normal documentation lifecycle,
   preserving unfinished work in `future.md`, repairing links and retaining a
   Git-history reference. Do not turn source completion into a publication claim.
6. Review and commit the remaining completed nonempty changes. Report the API
   and behavior delivered, verification, any remaining work, commit identifiers,
   and worktree state. Resolve supported failing checks within scope before
   claiming completion; a broad passing suite does not replace missing behavior.

Complete when the agreed target is implemented, verified and documented as one
coherent source state, with no new consumer restrictions or application policies.

## Open Questions

None. The feature roadmap resolves the public behavior and settings policy;
this plan selects the internal approach and validation ownership. Private helper
names, exact help prose, and test organization are ordinary implementation
choices within these decisions, not additional product gates.