/press

Use this skill after /cook has produced green implementation changes and before review or shipping.

Do not use it to implement broad new behavior. Press may add or strengthen tests and make tiny corrective fixes only when a test exposes a clear defect in the cooked scope.

--hard propagation

/press --hard (propagated from /cook --hard) is pass-through only. Press runs no gate. Hand --hard forward to /age at the handoff so it eventually reaches /cure, which is the only pipeline skill that fires the metacognitive vibecheck. See skills/hard-cheese/SKILL.md.

Flow

1. Read — load the spec or acceptance criteria and the cooked diff.
2. Map — for each changed behaviour, find the test(s) that cover it via cheez-search.
3. Gap analysis — identify weak assertions, missing boundaries, and uncovered integration seams. See references/gap-analysis.md for what counts as a gap and the priority order.
4. Add focused tests — observe red first when behaviour changes. Use cheez-write for precise edits.
5. Corrective fixes — only for defects the hardening tests expose. No new behaviour.
6. Run checks — narrowest useful tests, then relevant wider gates already in the project.
7. Report — write .cheese/press/<slug>.md (slug carried from /cook, or derived from branch/task) and print the path. Mark readiness: ready for /age, follow-up recommended, or blocked.
8. Hand off — in manual mode, prompt the next step via the shared handoff gate (see ## Handoff below); in --auto mode, chain forward per ### Auto mode.

Preferred tools and fallbacks

Code search, reading, and editing all go through the cheez-* skills (/cheez-search, /cheez-read, /cheez-write) — see those skills for tool selection rules. For coverage and test discovery, press uses cheez-search (callers via kind: "callers") and tilth_deps (cheez-search owns the routing).

Beyond cheez-* there are press-specific tools:

Need	Prefer	Fallback
Diff review	delta	plain git diff
Affected execution flows + risk scoring	code-review-graph: get_affected_flows_tool, get_impact_radius_tool	manual flow tracing from callers

Freshness: before the first code-review-graph query in a run, call build_or_update_graph_tool (and embed_graph_tool if you'll use semantic_search_nodes_tool to find shared-concept tests under divergent names). See /cheez-search for the full freshness contract and when semantic search beats tilth.

If optional tools are missing, press a narrower surface and state the residual risk.

Testing priority

1. Spec compliance: promised behavior has executable coverage.
2. Assertion strength: tests fail for wrong values, errors, or state.
3. Boundary behavior: empty, missing, malformed, minimal, and maximum inputs.
4. Integration seams: filesystem, subprocess, network, time, or dependency failure when in scope.
5. Happy path regression: the primary user path still passes.

Output

Write to .cheese/press/<slug>.md with a minimum handoff slug at the top so /ultracook and /cheese --continue can chain without re-parsing the report. The full report shape:

status: ok | halt: <one-line reason>
next: age | done
artifact: <path-if-any>
<one-line orientation: what press did — e.g., "added 4 boundary tests; no defects exposed">

# Press Report — <slug>

## Orientation
<one or two factual sentences about what press did this pass — the hardening added, the gaps closed, the readiness verdict. `/cheese --continue` surfaces the slug's orientation line to the user as "where you are", so press's orientation must describe press's own work, not duplicate cook's orientation.>

## Checks run
- <command>: <pass|fail|skipped with reason>

## Findings
| Severity | Category | Evidence | Recommendation |
| --- | --- | --- | --- |

## Coverage
- Spec coverage:
- Boundary coverage:
- Assertion strength:

## Readiness
<ready for /age | follow-up recommended | blocked>

## Next step
<ready for /age>:           /age <slug>           — review the cooked + pressed diff
<follow-up recommended>:    /age <slug>           — review-safe; documented follow-ups addressed after review
<blocked>:                  resolve blocking issues before proceeding

status: ok maps to readiness ready for /age or follow-up recommended (both are review-safe — the cooked contract is sound and every changed behaviour has a hardening test). status: halt: <reason> maps to readiness blocked with the reason filled in. next: age when readiness is ready for /age or follow-up recommended; next: done only when readiness is blocked so the orchestrator stops the chain.

Then print:

Press report: .cheese/press/<slug>.md
Next step:    /age <slug>                          (when ready for /age or follow-up recommended)
              blocked — resolve before continuing  (when blocked)

Handoff

Pipeline: culture → mold → cook → [press] → age → cure → ship

After the press report is on disk, ask via the shared handoff gate in ../../shared/handoff-gate.md. Lead each option with the verb (what the user wants to do next); the skill command (with any in-scope --hard propagation) is the backing detail. Default options:

• Review the diff (recommended when readiness is ready for /age or follow-up recommended) — /age <slug>. For follow-up recommended, the cooked contract is sound and every changed behaviour has a hardening test; documented follow-ups can be addressed after review.
• Stop — dispatch none; defer review (use this if you want to harden manually before /age, even though the contract is review-safe).

Pre-select Review the diff when readiness is ready for /age or follow-up recommended. If the report is blocked, do not pre-select anything; the user decides whether to fix or escalate. After a non-stop selection, run the selected command immediately.

Auto mode

When invoked with --auto (propagated from /cook --auto):

• Skip the handoff gate entirely.
• If readiness is ready for /age or follow-up recommended, invoke /age <slug> --auto directly. Both states mean the cooked contract is sound and every changed behaviour has a hardening test; follow-ups are documented and review-safe.
• If readiness is blocked, stop the auto chain and surface the press report to the user. blocked is reserved for cases where human judgement is genuinely required: a false premise on the cooked contract, an unfixable level-1/2 gap inside cooked scope, a changed behaviour press could not lock with a stable hardening test, or spinning wheels (three attempts at the same gap without green).

When invoked from /ultracook

/ultracook spawns press as a fresh-context sub-agent and owns the chain itself. When the spawn prompt explicitly says "for THIS PHASE ONLY" and "do not chain forward to the next phase," honour the override: write .cheese/press/<slug>.md (with the handoff slug at the top) and stop. Do not invoke /age <slug> --auto from inside the sub-agent regardless of readiness. The orchestrator reads the handoff slug and either chains to age (when next: age) or halts (when next: done).

Rules

• Do not weaken assertions.
• Do not broaden implementation beyond the cooked contract.
• Every changed behaviour in the cooked diff leaves press with an executable hardening test that would fail if the change regressed. Even on a 3-line off-by-one fix, press writes (or confirms cook wrote) a test like test_off_by_one_at_boundary that locks the fix in. If press cannot produce a stable hardening test for a changed behaviour (flaky seam, missing infrastructure, design decision required), readiness is blocked — never ready for /age or follow-up recommended.
• Cap iteration at three attempts per gap. Count test-edit + run cycles. On the third failed cycle on the same gap, mark readiness blocked with reason spinning: <gap-description> and surface the report. Do not loop indefinitely.
• Surface medium and high findings explicitly; summarize low findings.
• If the cooked diff or spec rests on a false premise (the contract is wrong, or the test surface is solving the wrong problem), stop and surface the premise before adding tests; do not harden the wrong angle.
• Apply the shared voice kernel (lives at skills/age/references/voice.md in this repo): lead the press report with the readiness verdict, flag residual risk as certain | speculating | don't know, agree when coverage is already sufficient without manufacturing tests.