Printing Press Polish

/printing-press-polish

Polish a generated CLI so it passes verification and is ready to publish.

The retro improves the Printing Press. Polish improves the generated CLI. This skill runs in a forked context (context: fork) so its diagnostic and fix loop doesn't pollute the caller — the diagnostic spam, fix iterations, and re-diagnose noise stay scoped to the polish session, and the caller receives a clean summary.

/printing-press-polish redfin
/printing-press-polish redfin-pp-cli
/printing-press-polish ~/printing-press/library/redfin

When to run

After any /printing-press generation, especially when:

• The shipcheck verdict is ship-with-gaps
• The verify pass rate is below 80%
• The scorecard is below 85
• You want the CLI publish-ready in one pass

Can also be run standalone on any CLI in ~/printing-press/library/.

Setup

PRESS_HOME="$HOME/printing-press"
PRESS_LIBRARY="$PRESS_HOME/library"

Public-library hint

If the user's request includes phrasing like "polish notion in the public library", "polish from the public library", or "polish the published cal-com" — and the named CLI is not in $PRESS_LIBRARY/<slug>/ — they're asking to polish a CLI that lives upstream but not locally. Polish runs against the internal library, so the right move is to import first.

Suggest: /printing-press-import <slug> to bring it in, then re-run polish. Don't try to polish a CLI that isn't in the internal library.

If the named CLI is already in $PRESS_LIBRARY/<slug>/, the "public library" phrasing is informational — just proceed with polish and let the divergence check (below) handle any drift.

Resolve CLI

The argument can be:

• A short name: redfin (looks up $PRESS_LIBRARY/redfin)
• A full name: redfin-pp-cli (strips suffix, looks up $PRESS_LIBRARY/redfin)
• A path: ~/printing-press/library/redfin (used directly)

Resolution order:

1. If the argument is an absolute or ~-prefixed path and exists, use it
2. Try $PRESS_LIBRARY/<arg> (exact match — works for slug like redfin)
3. If arg has -pp-cli suffix, strip it and try $PRESS_LIBRARY/<slug> (e.g., redfin-pp-cli → redfin)
4. Fuzzy search: ls $PRESS_LIBRARY/ | grep -i <arg> for close matches

Caller scenarios. Polish has two callers and they pass different argument forms:

• Standalone (user-invoked, /printing-press-polish redfin). The arg is a slug or binary name; resolution lands on $PRESS_LIBRARY/<slug>/. This is the published copy and the right target.
• Mid-pipeline (main printing-press skill Phase 5.5). The arg is $CLI_WORK_DIR — an absolute path to ~/printing-press/.runstate/.../runs/.../working/<api>-pp-cli/. Resolution must hit rule 1. Do not paraphrase this to the slug — Phase 5.5 fires before the working CLI is promoted, so $PRESS_LIBRARY/<slug>/ either doesn't exist or holds the prior run's stale CLI.

The lock-status check in the next code block is the safety net for the mid-pipeline scenario: if a build lock is held for this CLI (under either name form), polish refuses to run. printing-press lock normalizes slug ↔ binary-name internally, so the check works regardless of which form the basename produces.

If no match or multiple matches, present via AskUserQuestion. Show at most 4 matches sorted by modification time (most recent first) with human-friendly relative timestamps (e.g., "generated 2 hours ago").

CLI_DIR="<resolved path>"
CLI_NAME="$(basename "$CLI_DIR")"

# Check if there's an active build lock — polish edits would be overwritten
# when the running build promotes to library.
_lock_json=$(printing-press lock status --cli "$CLI_NAME" --json 2>/dev/null)
if echo "$_lock_json" | grep -q '"held".*true'; then
  if echo "$_lock_json" | grep -q '"stale".*true'; then
    echo "Warning: stale lock exists for $CLI_NAME (build may have crashed)."
    echo "Proceeding with polish. Run 'printing-press lock release --cli $CLI_NAME' to clear."
  else
    echo "An active build is in progress for $CLI_NAME."
    echo "Polish edits would be overwritten when the build promotes."
    echo "Wait for the build to finish, then run polish."
    exit 1
  fi
fi

# Verify it's a valid Go CLI
if [ ! -f "$CLI_DIR/go.mod" ]; then
  echo "Not a valid CLI directory: $CLI_DIR"
  exit 1
fi

echo "Polishing: $CLI_NAME"
echo "Location: $CLI_DIR"

Find spec and research dir

API_SLUG="${CLI_NAME%-pp-cli}"
SPEC_PATH=""
for f in "$PRESS_HOME/manuscripts/$API_SLUG"/*/research/*.yaml "$PRESS_HOME/manuscripts/$API_SLUG"/*/research/*.json "$PRESS_HOME/manuscripts/$CLI_NAME"/*/research/*.yaml "$PRESS_HOME/manuscripts/$CLI_NAME"/*/research/*.json; do
  if [ -f "$f" ]; then
    SPEC_PATH="$f"
    break
  fi
done

# Build the spec flag once. Empty when no spec was found — diagnostic
# commands accept a missing --spec and degrade gracefully.
SPEC_FLAG=""
if [ -n "$SPEC_PATH" ]; then
  SPEC_FLAG="--spec $SPEC_PATH"
fi

# Locate the research dir (parent of the spec's research/ folder, i.e.
# manuscripts/<api>/<run-id>/). dogfood's --research-dir triggers
# checkNovelFeatures, which writes novel_features_built back into
# research.json AND syncs the verified list into .printing-press.json.
# Without this flag, legacy CLIs whose manifest predates the
# novel_features schema fail publish-validate's transcendence gate.
RESEARCH_DIR=""
for d in "$PRESS_HOME/manuscripts/$API_SLUG"/*/research.json "$PRESS_HOME/manuscripts/$CLI_NAME"/*/research.json; do
  if [ -f "$d" ]; then
    RESEARCH_DIR="$(dirname "$d")"
    break
  fi
done

RESEARCH_FLAG=""
if [ -n "$RESEARCH_DIR" ]; then
  RESEARCH_FLAG="--research-dir $RESEARCH_DIR"
fi

Divergence check

Stop and run this step before Phase 1. Do not skip it. Do not proceed to diagnostics until you have completed the check and resolved any divergence.

The internal copy at $CLI_DIR can drift from the public library (mvanhorn/printing-press-library) copy if anyone edited the public repo directly after this CLI was last published. Polishing a stale internal copy and re-publishing later silently overwrites those public-only fixes — a real failure mode that shipped CLIs hit.

You must:

1. Locate the public library clone. Honor $PRINTING_PRESS_LIBRARY_PUBLIC if set; otherwise scan the user's filesystem however fits this platform. Validate every candidate by checking the git remote points at mvanhorn/printing-press-library — other directories may share the name (forks, accidental name collisions). If multiple valid clones exist, prefer the most recently modified; ask the user to disambiguate only if still unclear.

2. Locate this CLI inside the clone. find <clone>/library -type d -name "<api>-pp-cli" or equivalent.

3. Run diff -r <public-cli-dir> $CLI_DIR with these exclusions, all of which are expected to diverge after publish:

   • .printing-press-tools-polish.json (local ledger, not published)
   • go.mod and go.sum — publish rewrites the module path from <api>-pp-cli to github.com/mvanhorn/printing-press-library/library/<category>/<api>
   • All .go files where the only difference is the rewritten import path (the publish step propagates the new module path through every internal import). When inspecting .go diffs, scan for substantive changes — anything beyond the module-path prefix swap is real divergence.

   Concretely: diff -r --exclude=go.mod --exclude=go.sum --exclude=.printing-press-tools-polish.json <public-cli-dir> $CLI_DIR.

   Don't pass --exclude='<api>-pp-cli' or --exclude='<api>-pp-mcp' — those names match both the root-level binary files and the cmd/<api>-pp-cli/ and cmd/<api>-pp-mcp/ source directories. Excluding by binary name silently skips the entire cmd/ subtree, hiding real divergence in main.go. The "Only in $CLI_DIR: -pp-cli" line for the built binary is one row of expected output, not noise worth filtering at the cost of completeness.

4. Surface the result before continuing.

Outcomes:

• No clone found → user doesn't have public locally. State this explicitly ("public library not found locally; proceeding on internal as canonical") and continue.
• Clone found but doesn't contain this CLI → never published or under a different name. State this and continue.
• Found and diff is empty → in sync. State this and continue.
• Found and divergent → stop. Do not run Phase 1 diagnostics yet. List the divergent files for the user. Ask via AskUserQuestion: sync public→internal, or proceed without syncing. If the user picks sync, copy public's version of the divergent files into internal, then continue polish on the synced internal copy.

Before showing the sync prompt, check whether internal has files modified after its .printing-press.json timestamp (the user has been polishing locally without publishing). If yes, hedge the prompt explicitly: syncing will overwrite their pending local work. Let them decide whether to keep their local edits or pull public's.

After sync (or explicit skip), the rest of polish operates on $CLI_DIR as canonical. The eventual /printing-press-publish step pushes internal back to public; no second divergence check is needed there.

The check has run only when one of the four outcomes above is explicitly stated in your response. Silent omission counts as not having run it.

Phase 1: Baseline diagnostics

cd "$CLI_DIR"

# Build
go build -o "$CLI_NAME" ./cmd/"$CLI_NAME" 2>&1

# Diagnostics. SPEC_FLAG and RESEARCH_FLAG are set in the "Find spec
# and research dir" step above. RESEARCH_FLAG enables dogfood to
# verify novel features and sync them into .printing-press.json
# (required for publish-validate's transcendence gate).
printing-press dogfood --dir "$CLI_DIR" $SPEC_FLAG $RESEARCH_FLAG 2>&1
printing-press verify --dir "$CLI_DIR" $SPEC_FLAG --json 2>&1
printing-press workflow-verify --dir "$CLI_DIR" --json > /tmp/polish-workflow-verify.json 2>&1 || true
printing-press verify-skill --dir "$CLI_DIR" --json > /tmp/polish-verify-skill.json 2>&1 || true
printing-press publish validate --dir "$CLI_DIR" --json > /tmp/polish-publish-validate.json 2>&1 || true
# --live-check samples novel-feature outputs and populates
# live_check.features[].warnings (Wave B entity detection) — required for
# the "Output entity warnings" row below to have data to read.
printing-press scorecard --dir "$CLI_DIR" $SPEC_FLAG --live-check --json > /tmp/polish-scorecard.json 2>&1 || true
printing-press scorecard --dir "$CLI_DIR" $SPEC_FLAG 2>&1
printing-press tools-audit "$CLI_DIR" --json > /tmp/polish-tools-audit-before.json 2>&1 || true
go vet ./... 2>&1

verify-skill, workflow-verify, and publish-validate run alongside dogfood/verify/scorecard so polish catches the same class of failures the public-library CI catches. The publish-validate leg is a hard ship-gate: polish cannot recommend ship or ship-with-gaps while printing-press publish validate reports passed: false.

If Phase 1 baseline reveals the underlying CLI needs re-discovery — broken HTML/SSR extraction, sparse capture (fewer than 5 unique endpoints in the source manuscript), wrong endpoint shapes, missing GraphQL operation hashes, or any signal that the CLI was generated from incomplete capture — polish does not normally do browser capture itself, but the shared playbook at skills/printing-press/references/browser-sniff-capture.md covers all available capture backends including the Claude chrome-MCP (mcp__claude-in-chrome__*) and computer-use (mcp__computer-use__*) when the runtime exposes them. Read Step 1 (tool detection), Step 2c.5 (failure-recovery menu), and Step 2e (chrome-MCP capture playbook) of that reference before improvising. Re-discovery from polish is rare but real; when it happens, use the shared backends — do not invent a new capture flow.

Parse findings into categories:

Environmental failures vs. CLI defects. Some Phase 1 outputs surface failures that aren't real CLI bugs and should not block ship:

• scorecard --live-check reporting SQLITE_BUSY, network timeouts, 401 from a mock or expired token, or HTTP errors that depend on the test workspace's permissions/state — these are test-environment issues, not CLI defects.
• verify mock-harness flakes on commands with binary output (e.g., qr returning a PNG that the substring matcher can't validate) or commands with optional positional args where dry-run output legitimately doesn't contain the verify probe string.

Classify these as environmental in skipped_findings with the specific reason; do not spend Phase 2 cycles trying to "fix" them. The polish skill's ship logic already excludes live-check failures from gating, but the agent should still annotate them so reviewers can see they were considered and dismissed deliberately.

Phase 4.85 — Agentic output review (Wave B)

After the mechanical diagnostics above complete, invoke the printing-press-output-review sub-skill via the Skill tool. The sub-skill carries context: fork and owns the dispatch prompt, gate logic, and known blind spots — single source of truth shared with the main printing-press skill.

Skill(
  skill: "cli-printing-press:printing-press-output-review",
  args: "$CLI_DIR"
)

Parse the returned ---OUTPUT-REVIEW-RESULT--- block. status: WARN findings flow into the diagnostic categories above so Phase 2 fixes address both rule-based and plausibility issues. status: SKIP is informational — record but don't block.

Wave B gating applies: all findings are warnings, never blockers. Fix if obvious and cheap; document with a short comment if deferred.

Record baseline scores: scorecard total, verify pass rate, dogfood verdict, go vet issue count, output-review finding count.

Phase 2: Fix

Fix in priority order. After each priority level, update the lock heartbeat:

printing-press lock update --cli "$CLI_NAME" --phase polish 2>/dev/null

Runtime variant default checklist

If a polish fix adds or changes a runtime mode, data-source option, auth tier, transport, or other user-visible default, document this short checklist before selecting the default:

• User-visible default: which behavior users get without extra flags or config.
• Compatibility risk: whether existing commands, scripts, MCP tools, or stored config change behavior.
• Verification command: the exact command that proves the default and the non-default escape hatch both work.

Keep the checklist in the polish notes or result block. Skip it for ordinary bug fixes that do not change runtime variants or defaults.

Priority 0: MCP surface migration (legacy CLIs)

If Phase 1's dogfood reported MCP Surface: FAIL with a parity mismatch, the CLI was generated before the runtime cobratree walker existed and is still on the static internal/mcp/tools.go surface. The fix is mechanical:

printing-press mcp-sync "$CLI_DIR"

That migrates the MCP surface to the runtime walker, regenerates tools-manifest.json and internal/mcp/tools.go, and applies any mcp-descriptions.json overrides. Re-run dogfood after; the parity gate flips to PASS. This is a known migration path for every CLI generated before the cobratree landed; running it on a CLI already on the runtime walker is a no-op refresh.

Skip this priority on CLIs where dogfood's MCP gate is already passing.

Priority 1: Verify failures

For each command that fails verify dry-run or exec:

1. Read the command file
2. Find Args: cobra.ExactArgs(N) or similar constraint
3. Remove the Args: field
4. Add at the top of RunE:

   if len(args) == 0 {
       return cmd.Help()
   }
   

5. For commands needing 2+ args, use if len(args) < 2
6. Check for dry-run nil-data crashes and add guards:

   if flags.dryRun {
       return nil
   }
   

Priority 2: Dead code

1. For each dead function flagged by dogfood, grep all .go files to verify it's truly unused (not just its definition matching itself)
2. If truly unused: remove the function
3. If used by another helper: leave it (false positive)
4. After removal, remove unused imports
5. Delete stale files (promoted commands not registered in root.go)

Priority 3: CLI description and metadata

1. Read root command Short in internal/cli/root.go
2. If it contains boilerplate ("Reverse-engineered...", raw API title), rewrite: Pattern: "<Product> CLI with <capability-1>, <capability-2>, and <capability-3>"
3. Check commands for missing Example fields. Add realistic examples with domain-specific values.

Priority 4: README

Cardinal rule: run <cli> <cmd> --help for EVERY command you put in the README. Never guess flag names, argument formats, or valid values. If you write --start-time but the flag is --start, the README is wrong and users will get errors on their first try.

Source-of-truth files for rendered sections

Before editing README.md, SKILL.md, or .printing-press.json, identify whether the section is rendered from a source file. Dogfood and regeneration overwrite these rendered sections, so direct edits there are temporary and should be used only to inspect the current output.

Recommended loop for these rendered sections: edit the source field, re-run dogfood with --research-dir "$RESEARCH_DIR" or regenerate the CLI as appropriate, then run a second pass to confirm the rendered README/SKILL text stays fixed. If you edit README.md or SKILL.md directly in one of these sections, expect the next dogfood resync or regeneration to clobber the change.

To find the manuscript source:

PRESS_HOME="$HOME/printing-press"
API_SLUG="${CLI_NAME%-pp-cli}"
RESEARCH_JSON=""
for f in "$PRESS_HOME/manuscripts/$CLI_NAME"/*/research.json \
         "$PRESS_HOME/manuscripts/$API_SLUG"/*/research.json; do
  if [ -f "$f" ]; then RESEARCH_JSON="$f"; break; fi
done

If RESEARCH_JSON exists and a rendered section has bad prose, examples, or flag references, fix the corresponding field in that file first. For novel features, dogfood verifies research.json::novel_features[], writes the surviving set to research.json::novel_features_built[], and syncs README ## Unique Features, SKILL ## Unique Capabilities, .printing-press.json novel_features, and root help Highlights from that verified set.

Required sections (must be present and correct)

1. Title: "# CLI" — use the product's real name with correct casing/punctuation (e.g., "Cal.com" not "Cal Com")
2. Subtitle: one sentence describing what the CLI does for the user, matching the root Short field. NOT a description of the API.
3. Install: correct install command. Use the printing-press-library repo URL, not a per-CLI repo that doesn't exist.
4. Authentication: how to set <API>_API_KEY env var, where to get a key (link to the provider's settings page), self-hosted URL override if supported. Read config.go to find all env vars.
5. Quick Start: 3-5 commands someone will actually run first. Pick commands that are both most useful (what you'd run daily) and show the CLI's value (why install this over curl). Usually: doctor → sync → transcendence command (today, health) → search. Avoid raw list commands — they dump data without demonstrating why the CLI exists.
6. Commands: categorized table. Group by domain function (Scheduling, Analytics, Account, Utilities), not by implementation structure.
7. Output Formats: show --json, --select, --csv, --compact, --dry-run, --agent. Use a real command, not a placeholder.
8. Agent Usage: agent-native properties and exit codes.
9. Cookbook: 8-15 recipes using verified flag names from --help. Show the CLI's unique capabilities: transcendence commands, filters, SQL queries, pipes. Include at least one mutation example.
10. Health Check: show actual doctor output, not a placeholder.
11. Configuration: list ALL env vars from config.go with descriptions. Include config file path.
12. Troubleshooting: common errors mapped to exit codes with fixes.

Optional sections (add at your discretion)

• Rate Limits: if the API has documented limits
• Self-Hosting: if the CLI supports --api-url or BASE_URL override
• Pagination: if the API has notable pagination behavior
• Sources & Inspiration: credits to community projects (generated by the machine, preserve if present)

Priority 4.5: SKILL static-check failures (verify-skill)

Read /tmp/polish-verify-skill.json for the full finding list. Each finding has a check (flag-names, flag-commands, positional-args, unknown-command, or canonical-sections), a command (the path the SKILL claimed), and a detail describing the mismatch. Common shapes and fixes:

• flag-names — SKILL references --foo on a <cli> ... invocation but no command in internal/cli/*.go declares it. Either the example is wrong (fix the SKILL or remove the recipe) or the flag was deleted (decide if it should come back). Out of scope: flags on lines that invoke other tools (e.g. npx -y @mvanhorn/printing-press install <api> --cli-only, gh pr create --base ..., go install ...). The recipe-scoped flag-names check ignores those by design — never strip an external-tool flag to make verify-skill exit 0, and never replace the install instructions with a fabricated slash command. If the finding is firing on an external-tool flag anyway, that is a verify-skill bug, not a SKILL bug; report it instead of editing the SKILL.
• flag-commands — --foo is declared elsewhere but not on <cmd>. The flag exists somewhere but not on the command the SKILL invoked it on. Two fixes:
  1. If the flag is added via a shared helper like addXxxFlags(cmd, ...), inline the cmd.Flags().StringVar(...) declaration directly in the affected command's source file. The verify-skill grep cannot follow function-call indirection.
  2. If the SKILL example is genuinely wrong, fix the example to use a flag the command does declare.
• positional-args — got N positional args; Use: "<cmd> <arg>" expects M-M. The SKILL recipe passed N positional args but the command's Use: declares M required. Two fixes:
  1. If the command also accepts the value via a --flag, change Use: "cmd <arg>" to Use: "cmd [arg]" (square brackets = optional). Verify-skill correctly accepts --flag-only invocations against an optional positional.
  2. If the SKILL example is missing a required positional, fix the example.
• canonical-sections — install section drift: hand-edit detected in a generator-owned section. The ## Prerequisites: Install the CLI block has been edited away from what the generator would emit for this CLI today. Do not hand-edit the install section. It's templated from internal/generator/templates/skill.md.tmpl parameterized on (api_name, category, uses_browser_http_transport); any drift means an automation step or person modified text the machine owns. Resolve by regenerating the printed CLI (run printing-press regen against this directory, or for a published CLI, regenerate from the spec and re-publish). If the canonical text itself is wrong (e.g., a real change to the install instructions is needed), fix the template, not the printed CLI.

When editing other parts of SKILL.md, Read the affected section first and Read it again after the Edit. Edit replaces a literal string; if the surrounding context has drifted, a single Edit can graft a second copy of a block onto the first instead of replacing it.

After fixing, re-run printing-press verify-skill --dir "$CLI_DIR" and confirm exit 0 before moving on.

Priority 5: Remaining dogfood issues

• Path validity mismatches
• Auth protocol mismatches
• Example drift (examples referencing wrong commands)
• Data pipeline integrity issues

Priority 6: MCP tool quality

Your goal now is to ensure every MCP tool exposed by this CLI carries agent-grade descriptions and correct read/write classifications. Tool descriptions and classifications are how agents discover and decide whether to call a tool — thin descriptions and missing annotations directly degrade agent UX, and Phase 1's mechanical gates (verify, dogfood) do NOT catch this class of issue.

Stop and:

1. Run printing-press tools-audit "$CLI_DIR" --json to surface mechanical findings (empty Short, thin Short, missing mcp:read-only on read-shaped command names).
2. You must read references/tools-polish.md and follow its instructions to address the findings AND run a judgment pass over every command — regardless of whether the audit flagged it. The audit catches mechanical issues; description quality and borderline classification (read-only vs. local-write) always require agent reasoning. You must not skip this.
3. Accepting MCP-description findings carries a stricter contract. thin-mcp-description and empty-mcp-description accepts require three pre-decision fields (spec_source_material, target_description, gap_analysis) populated per finding. The binary rejects bulk accepts (>5 findings sharing one rationale) and runs that "complete" without lifting MCPDescriptionQuality. Fix via override or generator improvement is the expected path; accept is rare. See references/tools-polish.md "Marking a finding accepted" for the full contract.

Proceed to "After all fixes" only when the audit's summary line reads no pending findings with no incomplete: block — every gate (pre-decision fields, duplicate rationale, scorecard delta) passes.

After all fixes

go build -o "$CLI_NAME" ./cmd/"$CLI_NAME"
gofmt -w .

Phase 3: Re-diagnose

Re-run the diagnostic sweep on the fixed CLI:

printing-press dogfood --dir "$CLI_DIR" $SPEC_FLAG 2>&1
printing-press verify --dir "$CLI_DIR" $SPEC_FLAG --json 2>&1
printing-press workflow-verify --dir "$CLI_DIR" --json 2>&1
printing-press verify-skill --dir "$CLI_DIR" --json 2>&1
printing-press publish validate --dir "$CLI_DIR" --json 2>&1
printing-press scorecard --dir "$CLI_DIR" $SPEC_FLAG 2>&1
printing-press tools-audit "$CLI_DIR" 2>&1
go vet ./... 2>&1

Record the after scores. If verify-skill still has any severity=error findings, workflow-verify still reports workflow-fail, or publish-validate still reports passed: false, ship cannot fire (see ship logic below).

Ship logic

Compute the ship recommendation:

• ship: verify >= 80%, scorecard >= 75, no critical failures, AND verify-skill exits 0 (no SKILL/CLI mismatches), AND workflow-verify is not workflow-fail, AND publish-validate reports passed: true, AND tools-audit shows zero pending findings (every finding fixed or explicitly accepted with rationale). The SKILL/workflow/publish gates are hard requirements: a CLI that ships with a SKILL that lies about it (verify-skill findings) gives agents broken instructions; a CLI whose primary workflow fails verification has not actually shipped; a CLI that publish-validate rejects is not publishable.

• ship-with-gaps: verify >= 65%, scorecard >= 65, non-critical gaps remain, AND the SKILL/workflow/publish gates above hold, AND the README has a ## Known Gaps block that lists the user-facing gaps. Reserved for the rare case where a refactor or external-dependency blocker prevents a clean fix.

  README Known Gaps is mandatory for ship-with-gaps. The published library copy is what downstream users see; if the verdict claims gaps exist but the README hides them, downstream users meet a CLI that misbehaves with no disclosure. Before emitting ship_recommendation: ship-with-gaps:

  1. Read the CLI's README.md. If a ## Known Gaps section already exists (e.g., the main SKILL Phase 4 wrote it before polish ran), confirm it covers the user-facing items in remaining_issues. Add bullets for any newly surfaced user-facing gap polish discovered.

  2. If ## Known Gaps is missing, write it — placed after ## Quick Start (or before ## Usage) to mirror the ## Unique Features placement convention. One bullet per user-facing item from remaining_issues. Phrase from the user's perspective: what command misbehaves, what the workaround is. Example:

     ## Known Gaps
     
     - **`analytics export --csv`** returns truncated rows on workspaces with >10k events. Use `--json` and pipe to `jq` as a workaround until the underlying export endpoint is paginated.
     

  3. Filter remaining_issues for user-facing entries when populating the section. Internal items (verify drift on a deprecated flag, MCP description tuning, polish-internal notes) do not belong in the public Known Gaps. If the agent cannot identify any user-facing gap from remaining_issues, the verdict is ship, not ship-with-gaps.

  4. List each Known Gaps write/update in fixes_applied so the caller can surface that this happened.

  If polish cannot responsibly populate Known Gaps from the available evidence (e.g., remaining_issues is all internal jargon with no user-facing reading), downgrade the verdict to hold rather than ship without disclosure.

• hold: verify < 65% or scorecard < 65 or critical failures, OR verify-skill has unresolved findings, OR workflow-verify reports workflow-fail and the workflow is the CLI's primary value.

Push higher without gaming

The ship gates are a floor, not a ceiling. After they pass, look at scorecard dimensions still below max and ask whether each gap is real or structural:

1. Find the underlying deficit, not the score. The scorecard is a proxy for quality, not the goal itself. A README scoring 8/10 might be missing a Cookbook section or have outdated commands — that's a real, fixable gap. A mcp_surface_strategy scoring 2/10 on a 200-endpoint API might be flagging that the surface is mostly endpoint mirrors — also potentially fixable.
2. If there's a real, agent-grade improvement available, make it. Better description, missing flag doc, weak README section, an example that doesn't reflect actual usage. The CLI gets better and the score follows.
3. If the deficit is structural, document and accept. Some dimensions assume capabilities the CLI's domain doesn't have (a read-only API scored against write-workflow dimensions, a CLI with no auth scored on auth dimensions, a small API penalized on surface_strategy thresholds calibrated for large APIs). Note the reason in skipped_findings and move on.
4. Never add scaffolding to satisfy the scorer. Fake commands, fake tests, fake flags, or boilerplate prose written purely to nudge a number — those degrade the CLI to satisfy the proxy. The scorer is imperfect by design (the "scoring may be imperfect" caveat in AGENTS.md applies). Trust the underlying judgment, not the number.

MCP scorecard dims map to spec fields, not generator code

When mcp_token_efficiency, mcp_tool_design, mcp_remote_transport, or mcp_surface_strategy are below max, the fix is almost always a spec edit + regenerate (or regen-merge from a freshly-generated tree), not a generator-template change. Polish CAN address these — do not classify them as "feature add to a generator-owned file, retro candidate."

Recommended threshold: at >50 typed endpoints, default to recommending all four (transport, endpoint_tools=hidden, orchestration=code, intents for the headline workflows). Below 30, transport=[stdio, http] is the only zero-cost win. The full reference is docs/SPEC-EXTENSIONS.md.

After editing the spec, regenerate (or regen-merge the changes into the published library) so the new mcp: block reaches templates. Cobratree-walked novel commands continue to surface as MCP tools either way; they don't need spec changes.

Rule of thumb: if your fix would still be valuable if the scorecard didn't exist, do it. If the only motivation is "to push the score," don't.

Display delta and emit result block

Display the delta to the user, then emit the structured ---POLISH-RESULT--- block. The block lets calling skills (e.g., main printing-press SKILL.md Phase 5.5) parse the recommendation and scores reliably; the human-readable table above is for the user.

Polish Results for <CLI_NAME>:

                    Before    After     Delta
  Scorecard:        XX/100    XX/100    +N
  Verify:           XX%       XX%       +N%
  Tools-audit:      XX        XX        -N pending findings

Fixes applied:
  - <one-line description of each fix>

Skipped findings:
  - <finding>: <why you chose not to fix it>

Remaining issues:
  - <one-line description of each issue you tried to fix but couldn't>

---POLISH-RESULT---
scorecard_before: <N>
scorecard_after: <N>
verify_before: <N>
verify_after: <N>
dogfood_before: <PASS|FAIL>
dogfood_after: <PASS|FAIL>
govet_before: <N>
govet_after: <N>
tools_audit_before: <N pending>
tools_audit_after: <N pending>
publish_validate_before: <PASS|FAIL>
publish_validate_after: <PASS|FAIL>
fixes_applied:
- <one-line description of each fix>
skipped_findings:
- <finding>: <why you chose not to fix it>
remaining_issues:
- <one-line description of each issue you tried to fix but couldn't>
ship_recommendation: <ship|ship-with-gaps|hold>
further_polish_recommended: <yes|no>
further_polish_reasoning: <one sentence explaining the call>
---END-POLISH-RESULT---

The three lists serve different purposes:

• fixes_applied: what changed — the caller displays these
• skipped_findings: issues you found but deliberately did not fix, with reasoning (e.g., "verify classifies stale as read — scorer bug, not a CLI problem", "thin-short on version accepted as-is — accurate and brief"). The caller surfaces these so the user can decide whether to address them manually.
• remaining_issues: issues you tried to fix but couldn't resolve.

Picking further_polish_recommended

Your judgment, not a count of remaining_issues. Set yes when another polish invocation has a real chance of closing what's left:

• remaining_issues includes verify or dogfood failures you ran out of time on and a fresh pass with more attention per failure could plausibly resolve.
• The fixes you did land may have unblocked dependent issues you couldn't reach this pass.
• A SKILL/CLI mismatch needs a second look after this pass changed the source tree.

Set no when another invocation would re-tread the same ground:

• remaining_issues are decisions only the user can make (rename a flagship command, choose a default behavior, accept a structural trade-off).
• You already attempted the fix in two different ways this pass and both failed for the same reason.
• The blocker is external (API changed shape, rate-limited, missing credential) and not something a fresh polish run sees differently.
• remaining_issues is empty AND skipped_findings are all environmental or structural — there is nothing left for polish to do.

further_polish_reasoning is one sentence the caller surfaces verbatim. Make it specific ("verify failures on analytics export and report show looked closable but I gave up too early") rather than generic ("more polish might help"). Callers use this signal to decide whether to offer "Polish again" in their next prompt; a vague reason makes their prompt vague.

Publish Offer

Skip this entire section in mid-pipeline mode. Detect from $CLI_DIR: if the path is under .runstate/ (i.e., $PRESS_RUNSTATE/<scope>/runs/.../working/<api>-pp-cli/), polish is being called from main SKILL Phase 5.5 or hold-path "Polish to retry," and the working CLI has not been promoted to library yet. /printing-press-publish <slug> resolves to $PRESS_LIBRARY/<slug>/, which is either empty or holds a stale prior run — invoking publish here would either fail to resolve or ship the wrong copy. The parent skill owns the publish flow on that path; just emit the result block and return.

A simple check:

case "$CLI_DIR" in
  *.runstate/*) echo "mid-pipeline; skipping Publish Offe

…(truncated)