Text Intake

Text Intake (Stage 0 Orchestrator)

Resolve what text is being worked on and initialize its canonical workspace.

This skill is the Stage 0 owner for intake and disambiguation. It should answer:

• what exact work does the user mean?
• is this add mode or apply mode?
• which replay profile fits the request?
• what should the pipeline produce?
• what risks, blockers, and open questions need to be recorded now?

It does not own:

• source discovery
• extraction/cleaning/segmentation execution
• replay execution itself
• final promotion/import

Quick Status

Existing workspaces: !find ${LYCEUM_TEXTS_DIR:-output/texts} -mindepth 1 -maxdepth 1 -type d 2>/dev/null | wc -l Bootstrap script: !ls scripts/init_text_pipeline_workspace.go 2>/dev/null | wc -l Pipeline profiles defined: !grep -c "Profile[A-Z]" internal/textpipeline/textpipeline.go 2>/dev/null

Commands

• /text-intake init [work] [--mode add|apply] [--profile <profile>] — Create or refresh a canonical workspace
• /text-intake inspect [work] — Inspect likely mode, scope, and ambiguity before initialization
• /text-intake gap [work] — Compare requested outcome against current workspace/repo state and identify missing assets
• /text-intake status [work] — Summarize intake state, profile, blockers, and next stage

Target: $ARGUMENTS

Owned Responsibilities

Owns

• work disambiguation
• add vs apply mode selection
• replay profile selection
• success-criteria capture
• initial risk/blocker/open-question capture
• creation or refresh of manifest.json and initial state.json

Does not own

• source hunting itself
• extraction or cleaning execution
• replay execution
• final review/ship

⚠️ URN Convention (Critical)

Do NOT manually set english_edition_urn, generated_edition_urn, or greek_edition_urn in manifest.json.

The import script generates URNs containing "versified" which the reader requires for row view detection. Manually overriding these fields breaks the reader layout:

• The reader checks for "versified", "verse", or "gen-eng" in the English edition URN
• Without this convention, row view will not activate
• Users will see a broken/incomplete reading experience

What to do instead:

• Leave these fields blank in manifest.json
• The import script will auto-generate correct URNs at Stage 10
• Generated URNs follow the pattern: <work_urn>.workspace-versified-eng1

When you might be tempted to set them:

• Stage 6b (versification) — DON'T set generated_edition_urn
• Stage 10 (import) — DON'T override the auto-generated URNs
• External edition reuse — Use existing_edition_urn instead, not edition URN fields

The validation in scripts/import_workspace.go will warn if this convention is violated.

Current Repo Implementation

The current scaffold for this skill is:

• scripts/init_text_pipeline_workspace.go
• internal/textpipeline/textpipeline.go
• internal/textpipeline/textpipeline_test.go

Canonical init command

nix-shell -p go --run "go run ./scripts/init_text_pipeline_workspace.go \
  -work 'WORK NAME' \
  [-mode add|apply] \
  [-profile PROFILE] \
  [-edition-urn URN]"

Example: add mode

nix-shell -p go --run "go run ./scripts/init_text_pipeline_workspace.go \
  -work 'Sophocles Antigone'"

Example: apply mode

nix-shell -p go --run "go run ./scripts/init_text_pipeline_workspace.go \
  -work 'Meditations' \
  -mode apply \
  -profile interlinear \
  -edition-urn 'urn:cts:greekLit:tlg0562.tlg001.perseus-grc2'"

Workflows

/text-intake inspect

Use before initializing when the request is ambiguous or the existing text state is unclear.

Decide

1. exact work identity
2. ambiguity in author/title/recension
3. whether the request is add or apply
4. likely replay profile
5. expected target output
6. known blockers or open questions

Inspect using

• existing pipeline workspaces in $LYCEUM_TEXTS_DIR/<slug>/ (defaults to output/texts/)
• existing editions/alignment artifacts in the repo
• pipeline profile definitions in internal/textpipeline/textpipeline.go

/text-intake init

Create or refresh the canonical workspace.

Required decisions

• mode: add or apply
• profile: one of the defined replay profiles
• work: canonical request label / work name

Useful optional data to record

• existing edition URN
• target output label
• base Greek edition strategy
• target reference system
• success criteria
• risks
• blockers
• open questions

Profiles currently encoded

• add
• source-upgrade
• cleaning
• segmentation
• witness-upgrade
• versification
• transliteration
• interlinear
• treebank-enrichment
• reliability
• full-rehab

Expected outputs

$LYCEUM_TEXTS_DIR/<slug>/
├── manifest.json
├── state.json
├── provenance.md
├── sources/
├── raw/
├── extracted/
├── clean/
├── structured/
├── witnesses/
├── versification/
├── interlinear/
├── qa/
└── replay/stage-history.json

Note: LYCEUM_TEXTS_DIR defaults to output/texts for local development.

/text-intake gap

Use for apply-mode planning or when resuming a text.

Compare

• requested profile vs current stage states
• requested output vs existing artifacts
• known repo assets vs missing assets
• current blockers vs new request

Typical questions

• does the workspace already exist?
• which stages are already requested / done / invalidated?
• is there already an edition URN or prior artifact set to reuse?
• which next stage should run?

Useful checks

find ${LYCEUM_TEXTS_DIR:-output/texts} -mindepth 1 -maxdepth 2 | sort

nix-shell -p jq --run "jq '.' ${LYCEUM_TEXTS_DIR:-output/texts}/SLUG/manifest.json"

nix-shell -p jq --run "jq '.' ${LYCEUM_TEXTS_DIR:-output/texts}/SLUG/state.json"

/text-intake status

Summarize:

• workspace path
• mode and profile
• requested stages
• current stage
• blockers/open questions
• next recommended skill to run

Outputs

Canonical outputs

• $LYCEUM_TEXTS_DIR/<slug>/manifest.json
• initial state.json
• initialized workspace directory tree
• initial replay/stage-history.json
• intake notes in provenance.md and/or qa/final-review-pack.md

Current source of truth

• internal/textpipeline/textpipeline.go

Verification Contract

This skill follows the Stage 0 contract from docs/text-pipeline-skill-verification-2026-03-13.md.

Verify

• requested work was resolved correctly
• add vs apply mode is correct
• replay profile/stage scope is correct
• manifest fields are complete
• initial state marks only the intended stages as in scope

Minimum evidence

• $LYCEUM_TEXTS_DIR/<slug>/manifest.json
• $LYCEUM_TEXTS_DIR/<slug>/state.json
• intake notes with risks/blockers/open questions

Pass criteria

• manifest identifies the intended work/edition target
• mode and profile match the request
• requested stages match the selected profile
• required manifest fields are non-empty
• state initializes the correct stages without accidental extra scope

Failure examples

• wrong work chosen for ambiguous title
• apply request initialized as full add
• wrong replay profile selected
• workspace exists but manifest/state do not match the request

Required next steps

After a successful intake run, the next owner is usually:

• source-hunt for add mode or source-oriented apply work
• text-replay when the user asked for a targeted replay profile

Verification

After completing this stage, run the automated verification script:

bash scripts/verify_stage_0.sh "${SLUG}"

Exit codes: 0=PASS (advance), 1=FAIL (block), 2=WARN (advance with notes). The orchestrator runs this automatically; when executing manually, check the output for [FAIL] or [WARN] lines.

Key Files