@harness-lab/cli 0.2.9 → 0.3.1

package/README.md

@@ -14,6 +14,7 @@ Current shipped scope:
 - `harness workshop status`
 - `harness workshop list-instances`
 - `harness workshop show-instance`
+- `harness workshop participant-access`
 - `harness workshop create-instance`
 - `harness workshop update-instance`
 - `harness workshop reset-instance`
@@ -130,6 +131,9 @@ harness workshop select-instance sample-workshop-demo-orbit
 harness workshop current-instance
 harness workshop status
 harness workshop show-instance sample-workshop-demo-orbit
+harness workshop participant-access
+harness workshop participant-access --rotate
+harness workshop participant-access --rotate --code orbit7-bridge4-shift2
 harness workshop create-instance sample-workshop-demo-orbit --event-title "Sample Workshop Demo"
 harness workshop update-instance --room-name Orbit
 harness workshop reset-instance --template-id blueprint-default
@@ -148,6 +152,7 @@ Targeting model:
 - `harness workshop current-instance` reports the stored target and resolves its current server state
 - `harness workshop status` and `harness workshop phase set <phase-id>` use the selected instance when present, otherwise they fall back to deployment default behavior
 - `harness workshop show-instance`, `update-instance`, `reset-instance`, `prepare`, and `remove-instance` accept an explicit `<instance-id>` but may also use the stored selection as a fallback
+- `harness workshop participant-access` accepts an explicit `<instance-id>` but may also use the stored selection as a fallback
 - `harness workshop select-instance --clear` removes the stored selection
 - `HARNESS_WORKSHOP_INSTANCE_ID` remains an environment fallback when no local selection is stored
 

package/assets/workshop-bundle/SKILL.md

@@ -243,6 +243,12 @@ Prefer invoking `harness workshop show-instance <instance-id>` over raw API scri
 Use this when the facilitator needs the full record for one instance rather than the deployment-default runtime status returned by `workshop facilitator status`.
 If a current instance is already selected, the CLI may omit `<instance-id>` and use the stored target.
 
+### `workshop facilitator participant-access`
+
+Inspect or rotate the shared participant event code for the current workshop instance.
+Prefer invoking `harness --json workshop participant-access` for inspection and `harness --json workshop participant-access --rotate` to issue a fresh code.
+If the current raw code is no longer recoverable from the hash-only runtime store, issue a new code instead of guessing.
+
 ### `workshop facilitator grant <email> <role>`
 
 Grant a Neon Auth user access to the current workshop instance.
@@ -320,6 +326,14 @@ Rules for rich scenes:
 
 Archive the current workshop instance with optional notes.
 
+### `workshop facilitator learnings`
+
+Query the cross-cohort learnings log to review rotation signals from past and current workshops.
+Prefer invoking `harness --json workshop learnings` for machine-readable output.
+Supports flags: `--tag TAG`, `--instance ID`, `--cohort NAME`, `--limit N` (default 20).
+When the facilitator asks for rotation signals, captured observations, or what happened during past handoffs, use this command.
+If the learnings log is empty, say so and suggest capturing the first signal using the rotation capture panel in the facilitator dashboard.
+
 ### `workshop analyze`
 
 Review the team's repo against the handoff criteria in `workshop-skill/analyze-checklist.md`.
@@ -349,10 +363,12 @@ Relevant local files:
 - `workshop-skill/follow-up-package.md`
 - `workshop-skill/facilitator.md`
 - `materials/participant-resource-kit.md`
+- `materials/coaching-codex.md`
 - `docs/learner-resource-kit.md`
 - `docs/learner-reference-gallery.md`
 - `content/project-briefs/*.md`
 - `content/challenge-cards/deck.md`
+- `content/codex-craft.md`
 - `workshop-blueprint/*`
 - `docs/workshop-event-context-contract.md`
 - `docs/harness-cli-foundation.md`

package/assets/workshop-bundle/bundle-manifest.json

@@ -1,8 +1,8 @@
 {
   "manifestVersion": 1,
   "bundleName": "harness-lab-workshop",
-  "bundleVersion": "0.2.9",
-  "contentHash": "671837951cf6679bea1d517c900927aab5d457806695e1a6da25574b8573f325",
+  "bundleVersion": "0.3.1",
+  "contentHash": "d52937fa7d980cdd71345b725b68bc934e939bc5b48cabd4d5bcfb55da03362f",
   "files": [
     {
       "path": "content/challenge-cards/.gitkeep",
@@ -10,19 +10,19 @@
     },
     {
       "path": "content/challenge-cards/deck.md",
-      "sha256": "c2828eb5258031351d6d5abd3388b26b0894a9122e88e3203efaccbfd95a3e82"
+      "sha256": "1c8b4e3a6ab987dc2e9e5418a32a84b8dd3485b8ee0d7d5dc9dc01394f57e3cc"
     },
     {
       "path": "content/challenge-cards/locales/en/deck.md",
-      "sha256": "6b6a3882ef31b3bca9f1f1ed43ef4fb2d8340af9879fee4261922fb55f6cd921"
+      "sha256": "f519c2dfa66801a90e1dbf8833e73bd03fbb001de9766286f9627750456a0b9b"
     },
     {
       "path": "content/challenge-cards/print-spec.md",
-      "sha256": "be7ecfdae9ada4145272f12e2d6018ccde7cfaeb9f48feb58aca7da7236815c0"
+      "sha256": "0c6bc2953d6296b8901705d938a29c1ef859490e53cc265fe401bb9585611e16"
     },
     {
-      "path": "content/czech-editorial-review-checklist.md",
-      "sha256": "23539ba8b2c6ed193cfadc7ad0f0f75e99b2926ebff5beade5df9789e6071fcf"
+      "path": "content/codex-craft.md",
+      "sha256": "d75b15cf97c4ab7f4918e09fff0efbe204d141f7127ca09d9a446a128c8efaaf"
     },
     {
       "path": "content/facilitation/.gitkeep",
@@ -30,11 +30,11 @@
     },
     {
       "path": "content/facilitation/codex-setup-verification.md",
-      "sha256": "0d7e36a8f2b8c68939997b117f1b5cf1538a5b9e27f2d18948a1ac407543de8e"
+      "sha256": "d137a5e729a1d443713a9e8431bffe169a1058d3e64c298688a6b7924fbdf3a8"
     },
     {
       "path": "content/facilitation/master-guide.md",
-      "sha256": "845adbcf9284cedd589d461b40ae1cd336d816b92bcc239df4574148a57b263a"
+      "sha256": "b5e107e88ab335aaa8870f51cf9c8b8dadb619d3b0f61dab087dacdcdef4c136"
     },
     {
       "path": "content/project-briefs/.gitkeep",
@@ -42,15 +42,15 @@
     },
     {
       "path": "content/project-briefs/code-review-helper.md",
-      "sha256": "3e174ff0d097fe132088774042f9db58c9c2494613d66950bdd423116bd7986d"
+      "sha256": "96bbbef60ab0ff93ad4aa3858852886329139a25ce9f375bc4cc57ad706b7c98"
     },
     {
       "path": "content/project-briefs/devtoolbox-cli.md",
-      "sha256": "867ba1b28413fcf43a95cc247ec07656f31b2fdc7bf4512fd5b4316a31f0e27b"
+      "sha256": "ce0b09a5e77be19f8ddb2ba6a43137c3cc2a415512d247971186e846dd018888"
     },
     {
       "path": "content/project-briefs/doc-generator.md",
-      "sha256": "3f3bea866c68a927916d90fa3e387ceccdd174290944bc83db7b89d43f3ba7a0"
+      "sha256": "48b8298be1e14e962055ac6d157be2728aa96fc71109e7ee833a6c56dab64731"
     },
     {
       "path": "content/project-briefs/locales/en/code-review-helper.md",
@@ -58,35 +58,27 @@
     },
     {
       "path": "content/project-briefs/locales/en/devtoolbox-cli.md",
-      "sha256": "cb7a92f20fa0565611feb9f24eb65aafc4a7b151a32de42c93d33b93049f161e"
+      "sha256": "3dd5e35524170feda7fa5e73d6b331d1bed88dd8c5966525d361fd1152802cb0"
     },
     {
       "path": "content/project-briefs/locales/en/doc-generator.md",
-      "sha256": "5ff49a57e51c030b34385709ab7519a5d8e19cd98a1e3a5b26603a7eb59c02e3"
+      "sha256": "b9f5b61dd21d867dc231bc5cf29ef1f493a9a208f55fc0f840b22302cd6890de"
     },
     {
       "path": "content/project-briefs/locales/en/metrics-dashboard.md",
-      "sha256": "ff6b429001ea4ab2d9c23fbdfe02c5daaaf413d52609f9dabc365902f2ad063a"
+      "sha256": "6ae0a048ba0692f9a4b88f4fce90be7ea9cf8a28f551b59870f4862d6e5785a5"
     },
     {
       "path": "content/project-briefs/locales/en/standup-bot.md",
-      "sha256": "bc3972786dac4298607c63e15262cf5330c7e7fcf754e5df64e8fae081f5b44b"
+      "sha256": "bd9aff3ac09e4866b10e90685e348e041dfe8cdbd8a65c17871f5d3b5f2240d0"
     },
     {
       "path": "content/project-briefs/metrics-dashboard.md",
-      "sha256": "a678058b6515fec4acba5d288bb71a725dfe9e75320efa11a475d63492db14f6"
+      "sha256": "40c0d09c02fb20c97c0f99088641de20fa966b17089f6f7701a7808814b31c55"
     },
     {
       "path": "content/project-briefs/standup-bot.md",
-      "sha256": "bf3af5bd1c72b5d4f3f5639c82d8e28dfbd4b470b0181833a2fff6e12eb99221"
-    },
-    {
-      "path": "content/style-examples.md",
-      "sha256": "c8e62493e119995fca1c738584b864506bc838203b1adad008c37b6f6ec57c50"
-    },
-    {
-      "path": "content/style-guide.md",
-      "sha256": "ec9a49e040efbc6c365eb264bd3452fdf1906f141f358ff28d2e0e3360648d28"
+      "sha256": "df9abc9d7bc8283d30d1c09fb412f9e072a980782c0b622747b13744b2a81c84"
     },
     {
       "path": "content/talks/.gitkeep",
@@ -94,15 +86,15 @@
     },
     {
       "path": "content/talks/codex-demo-script.md",
-      "sha256": "4f40188a8d3ccaddf44b89b39178c7e9cfb53298a8803815f91dce204181a9c2"
+      "sha256": "d5ff639fde21e47606a81aeae6e2b135d2063c073a8dd5717a0386546b9b77cc"
     },
     {
       "path": "content/talks/context-is-king.md",
-      "sha256": "0bb48497b458474a546c351118ad455c00ad997c9efc0e227b388f81dd0b55e3"
+      "sha256": "31fb6a02037c50e0b3fa2ae4bc39f658924bd164279fb490fc192d604763ff9f"
     },
     {
       "path": "docs/harness-cli-foundation.md",
-      "sha256": "cff448e5c1ee4e7129724e4f8294add35cd1efb621c83182b341a1b7e5a16b86"
+      "sha256": "10311d3caf8cec304d5d1694eac66623f06c100990d6f5c4d1dd1f4ba8a24bd8"
     },
     {
       "path": "docs/learner-reference-gallery.md",
@@ -110,7 +102,7 @@
     },
     {
       "path": "docs/learner-resource-kit.md",
-      "sha256": "c1c35a7896a835a49599688993f9782ad09e4f1c012b7c2b8d60b90b826eab1b"
+      "sha256": "1bfad4ee7514b76f9ad6943157224b345359588748ffffb7b93f2f8b70c8af06"
     },
     {
       "path": "docs/locales/en/learner-reference-gallery.md",
@@ -124,17 +116,21 @@
       "path": "docs/workshop-event-context-contract.md",
       "sha256": "a7706a398ea5250537fb34323c4371ea2a7db15e477aec3a7893e6089fcd281f"
     },
+    {
+      "path": "materials/coaching-codex.md",
+      "sha256": "2390be079acde89b16380cdd0904fac2cbefcb9aa11c1fe256330b2dc2362ba9"
+    },
     {
       "path": "materials/locales/en/participant-resource-kit.md",
-      "sha256": "1d8b2fbf019db86c56db9cd001965f97985461c79cd615cd861562f336b175fb"
+      "sha256": "02c7d6bde0c0662fdfdd0a8baed8c93f2a3c722c1a59336e4862f0d8f8490fbe"
     },
     {
       "path": "materials/participant-resource-kit.md",
-      "sha256": "19564e2e1156f60654e07a1c965f87668a8cb78cab5f9d06b1e1127ce812f13c"
+      "sha256": "d7cd7db05429faf7af570e33401cb5788d663e06c984f193596c14d2d8a12a5b"
     },
     {
       "path": "SKILL.md",
-      "sha256": "d1e242a857f61bc2fae29c74a9e0efbd1ba0ed5beb25087dbaeffabc5f53b4a5"
+      "sha256": "0188da9de8f517a09dc4d0e2eb14f395aa465048253bf0c719995e8a232a33e2"
     },
     {
       "path": "workshop-blueprint/agenda.json",
@@ -146,11 +142,7 @@
     },
     {
       "path": "workshop-blueprint/day-structure.md",
-      "sha256": "ffecae258211cef2430112ca25e1ff59658cac729ea2a07b9cac48e9a2a6244e"
-    },
-    {
-      "path": "workshop-blueprint/edit-boundaries.md",
-      "sha256": "d5a2b6039ff9d3acc5b3a58698289817a29d52af4058034f00bc188db6f5d84d"
+      "sha256": "22d6fec0422751439bd10049028ee12d140599707dbc6df248b4ff8517ee0c15"
     },
     {
       "path": "workshop-blueprint/operator-guide.md",
@@ -158,7 +150,7 @@
     },
     {
       "path": "workshop-blueprint/README.md",
-      "sha256": "badb477ecfef7df163ec41c72c39e04bcde2a5a723c1b01bf2e91ed38924e903"
+      "sha256": "e5cc8a87576988ce724b7ffacd86a4276aa6733bd4b9037d1f560a58da048308"
     },
     {
       "path": "workshop-blueprint/teaching-spine.md",
@@ -170,63 +162,63 @@
     },
     {
       "path": "workshop-skill/analyze-checklist.md",
-      "sha256": "b8ec5bb13a5ec25f0d3b8cbd60295c0b290aaae195d833956cd906092efc945d"
+      "sha256": "b1b5af58fc18f8400677a964eb3c1970015284d63101683fc8232bbea6cf8871"
     },
     {
       "path": "workshop-skill/closing-skill.md",
-      "sha256": "e83e381216fad93356b94da89ca2525e01978d13005f2b0f11cf5ed32382971a"
+      "sha256": "db23689de38b1ebfc2d9dde5387c5386a774cc6e90ab3e6c95f7749601d6abb7"
     },
     {
       "path": "workshop-skill/commands.md",
-      "sha256": "932305291e1fc943308f218a5d2714f43a2fea471f4702fe5c94192ed3cbc1e8"
+      "sha256": "d0af2bcc53c9752be3b013da002c22d4a8c7a09dea1995075b82550947141736"
     },
     {
       "path": "workshop-skill/facilitator.md",
-      "sha256": "525fc9fdad22cf42b47854ff9ad90579d728daf2c4ac484aba13f93193eb56d3"
+      "sha256": "b8c9e1c01ddb314e45d8d2c3a7460bac53a48ba2321eaf6518e4b08436d88895"
     },
     {
       "path": "workshop-skill/follow-up-package.md",
-      "sha256": "a8c9e847ac1c7dd03bc044ab4d82f8ce18f83c48b665aee3723e64487f845963"
+      "sha256": "74b8232f54b663d8ba9a97485c898ef0e8cffd5b5423831627ee7fe94d1cca44"
     },
     {
       "path": "workshop-skill/install.md",
-      "sha256": "2bb08f93523348de834d1e7004ef4603905c8d036247037e2111153cd0fea76b"
+      "sha256": "4e356194b0dbe114f71ec3661e1b4a4c323cdaa5efedac977f28629b4db66772"
     },
     {
       "path": "workshop-skill/locales/en/commands.md",
-      "sha256": "17856fa19a44c3336709186cba171505db4b503d11bc16ad2fbc49f67d6c78a0"
+      "sha256": "596af69cff67b770c284b4485c94bc39b511a60f3ff1b2b34f91f2c58a9f8b03"
     },
     {
       "path": "workshop-skill/locales/en/follow-up-package.md",
-      "sha256": "c3d4b0057ddc7a2a50713b6a01e7a3564b4e8668118464e19f932d2baa7a81a0"
+      "sha256": "f941c29394f143b9e78d2d54ffb3e0643e421c310329ead784a13e788c8d8caf"
     },
     {
       "path": "workshop-skill/locales/en/recap.md",
-      "sha256": "68141bac34417fbc8669ebd2e998658dcb66a2a8d1df6cf8ac8d2f449baacdae"
+      "sha256": "58329809e3400599a3b187f89dcae4aac939737b786389ed6c893540aa8e8249"
     },
     {
       "path": "workshop-skill/locales/en/reference.md",
-      "sha256": "5269750632ab577f9373cd84ae57741b11364cfbd9a05d6ac7dd8de555236b53"
+      "sha256": "2e941f974ae86c79131770d40302eec0e7968e217f36d8ff166065c7819c47b3"
     },
     {
       "path": "workshop-skill/locales/en/setup.md",
-      "sha256": "b5efeee9bb2ee1ada3ab2c86ae47c63afd51534fe58a849f0584384737150375"
+      "sha256": "3b77f4ca8d7eedf6d2ad01635c6727fd50fe658520b93f0bb1ae9c605ad62c31"
     },
     {
       "path": "workshop-skill/recap.md",
-      "sha256": "9310f61f5adcace4c97b249d0f27e3a423b5b3a278ec6f2440bf7a66d04a3058"
+      "sha256": "fa242bbb17d5305fda2b3c08a8da14744f60f6895985bbbbb1604c4c1ad4c8b4"
     },
     {
       "path": "workshop-skill/reference.md",
-      "sha256": "2023dabca1aba04a539152131e8cec2eb832c0e1107e2bb808bd8b296159bbdb"
+      "sha256": "fe7d1c5f32d2a88b4673f565fda8bb389c4fe3b2a50aad5e2a9fafef7f75d658"
     },
     {
       "path": "workshop-skill/setup.md",
-      "sha256": "ef67c05751f2c48182b834314e411fc912d0ec99f7cb28fab540534e2a07096c"
+      "sha256": "8a9a66372b49d504b5df45e0df3eac2835ef02ce79b7db1078fc85a928a36d5d"
     },
     {
       "path": "workshop-skill/template-agents.md",
-      "sha256": "07dd6ee5a017a0af01bf3bdcc465f06a9d8701851a413e805d29cbb22a7d991a"
+      "sha256": "a8647455ac66b79107d120258591e5e96cef394ed1b2acdb8a2ec4669469c5fe"
     }
   ]
 }

package/assets/workshop-bundle/content/challenge-cards/deck.md

@@ -1,38 +1,40 @@
 # Challenge Cards
 
-Karty nejsou body navíc. Jsou to malé zásahy, které zlepšují způsob práce s agentem i kvalitu handoffu.
+Karty nejsou body navíc. Jsou to malé zásahy, které zlepšují způsob práce s agentem i kvalitu handoffu.
 
-## Context Engineering
+## Před obědem: postavte pracovní systém
 
-- `Vytvořte AGENTS.md jako mapu` — sepište cíl, build/test příkazy, trvalá pravidla a kam má další tým sáhnout jako první.
+- `Vytvořte AGENTS.md jako mapu` — sepište cíl, build/test příkazy, trvalá pravidla a kam má další tým sáhnout jako první.
 - `Přidejte build/test příkazy` — agent musí umět ověřit výsledek bez ručního dovysvětlování.
-- `Napište skill pro code review` — formalizujte jednu review rutinu, kterou by uměl použít i cizí tým.
+- `Napište skill pro code review` — formalizujte jednu review rutinu, kterou by uměl použít i cizí tým.
 - `Přesuňte pravidlo z hovoru do repa` — vše, co už tým řekl dvakrát nahlas, převeďte do `AGENTS.md`, README, runbooku nebo testu.
 
-## Workflow
+## Po rotaci: opravte signál, ne jen feature
 
-- `Použijte /plan před kódováním` — ukažte, z jakého plánu tým vycházel, co z něj opravdu plní a jaký je další bezpečný krok.
-- `Rozdělte práci do více vláken` — zkuste dvě nezávislé linie práce a jednoho člověka na integraci.
+- `Použijte /plan před kódováním` — ukažte, z jakého plánu kroků tým vycházel, co z něj opravdu plní a jaký je další bezpečný krok.
+- `Rozdělte práci do více vláken` — zkuste dvě nezávislé linie práce a jednoho člověka na integraci.
 - `Delegujte úkol a vraťte se ke kontrole za 10 minut` — neskákejte agentovi do každého kroku, kontrolujte až výsledek.
-- `Přidejte nejmenší užitečné ověření` — vytvořte RED test, tracer bullet nebo jednoduchý browser check dřív, než agent dostane víc autonomie.
+- `Přidejte nejmenší užitečné ověření` — zapište done criteria jako spustitelný check (unit test, tracer bullet nebo jednoduchý browser check) dřív, než agent dostane víc autonomie.
+- `Napište diagnózu po handoffu` — pojmenujte, co pomohlo, co chybělo a jaký je další safe move.
+- `Opravte jeden slabý signál v repu` — README, AGENTS.md, plán, runbook nebo check.
 
 ## Advanced
 
-- `Spusťte 2 paralelní Codex sessions` — rozdělte problém na dvě nezávislé části a porovnejte výstupy.
-- `Vytvořte runbook pro deployment` — i kdyby deploy zůstal jen simulovaný.
-- `Napište AGENTS.md pro podsložku` — ukažte, že kontext může být globální i lokální.
-- `Zaveďte garbage collection` — najděte jednu opakující se formu chaosu a proměňte ji v check, template nebo pravidlo.
+- `Spusťte 2 paralelní Codex sessions` — rozdělte problém na dvě nezávislé části a porovnejte výstupy.
+- `Vytvořte runbook pro deployment` — i kdyby deploy zůstal jen simulovaný.
+- `Napište AGENTS.md pro podsložku` — ukažte, že kontext může být globální i lokální.
+- `Zaveďte garbage collection` — najděte jednu opakující se formu chaosu a proměňte ji v check, template nebo pravidlo.
 
 ## Meta
 
 - `Přesuňte trvalé pravidlo z promptu do AGENTS.md`.
 - `Přidejte sekci Done When ke každému tasku`.
 - `Napište README pro tým po rotaci, ne pro sebe`.
-- `Zapište, co je skutečně ověřené` — odlište hotové, rozpracované a jen předpokládané.
+- `Zapište, co je skutečně ověřené` — odlište hotové, rozpracované a jen předpokládané.
 
-## Jak s kartami pracovat
+## Jak s kartami pracovat
 
-- Před obědem má každý tým splnit aspoň jednu kartu z části `Context Engineering`.
-- Před rotací má být v repu dohledatelné, co bylo opravdu ověřeno a jaký je další bezpečný krok.
-- Po rotaci má každý tým splnit aspoň jednu kartu z `Workflow`.
+- Před obědem má každý tým splnit aspoň jednu kartu z části `Před obědem: postavte pracovní systém`.
+- Před rotací má být v repu dohledatelné, co bylo opravdu ověřeno a jaký je další bezpečný krok.
+- Po rotaci má každý tým splnit aspoň jednu kartu z `Po rotaci: opravte signál, ne jen feature`.
 - Ostatní karty jsou dobrovolné. Berte je jako stretch cíle nebo inspiraci, když nevíte, co zlepšit dál.

package/assets/workshop-bundle/content/challenge-cards/locales/en/deck.md

@@ -2,19 +2,21 @@
 
 These cards are not bonus points. They are small interventions that improve both the way of working with the agent and the quality of the handoff.
 
-## Context Engineering
+## Before lunch: build the working system
 
 - `Create AGENTS.md as a map` - write down the goal, build and test commands, durable rules, and where the next team should look first.
 - `Add build/test commands` - the agent must be able to verify the result without manual backfilling.
 - `Write a code review skill` - formalize one review routine that another team could use too.
 - `Move one rule from conversation into the repo` - anything the team has said out loud twice should be turned into `AGENTS.md`, README, a runbook, or a test.
 
-## Workflow
+## After rotation: fix the signal, not only the feature
 
 - `Use /plan before coding` - show which plan the team is working from, what is actually being executed, and what the next safe move is.
 - `Split the work into multiple threads` - try two independent work streams and one person on integration.
 - `Delegate a task and come back to check it in 10 minutes` - do not jump into every agent step; inspect the outcome instead.
-- `Add the smallest useful verification` - create a RED test, tracer bullet, or simple browser check before the agent gets more autonomy.
+- `Add the smallest useful verification` - write the done criteria as an executable check (unit test, tracer bullet, or simple browser check) before the agent gets more autonomy.
+- `Write the post-handoff diagnosis` - name what helped, what was missing, and what the next safe move is.
+- `Fix one weak repo signal` - README, AGENTS.md, plan, runbook, or check.
 
 ## Advanced
 
@@ -32,7 +34,7 @@ These cards are not bonus points. They are small interventions that improve both
 
 ## How to use the cards
 
-- Before lunch, every team should complete at least one `Context Engineering` card.
+- Before lunch, every team should complete at least one `Before lunch: build the working system` card.
 - Before rotation, the repo should clearly show what was actually verified and what the next safe move is.
-- After rotation, every team should complete at least one `Workflow` card.
+- After rotation, every team should complete at least one `After rotation: fix the signal, not only the feature` card.
 - The other cards are optional. Treat them as stretch goals or prompts for what to improve next.

package/assets/workshop-bundle/content/challenge-cards/print-spec.md

@@ -17,7 +17,7 @@
 
 1. nahoře název karty
 2. pod ním kategorie
-3. 2-3 věty zadání
+3. 2–3 věty zadání
 4. dole malý štítek:
    - `Před obědem`
    - `Po rotaci`

package/assets/workshop-bundle/content/codex-craft.md

@@ -0,0 +1,190 @@
+# Codex Craft
+
+> Tool-specific fluency to sit on top of the agent-agnostic harness method.
+
+Harness Lab teaches a method that transfers across coding agents: Codex, pi, Claude Code, Cursor, Aider. The method is agent-agnostic by design. This document is the other half: the **Codex-specific craft** that the method assumes you already know, but that nobody teaches you in order.
+
+If something in this doc contradicts the live Codex documentation, trust the live docs. This doc is a teaching artifact, not a specification. Last verified against Codex CLI in April 2026 — re-verify before each cohort.
+
+---
+
+## 1. What the harness actually is in Codex
+
+In Codex, the "harness" is the union of four things:
+
+1. **The repo context the model can see.** `AGENTS.md`, files you name, files it reads on its own, the diff you built up this session.
+2. **The tool affordances the model can reach.** Shell access, file edits, network calls, Playwright probes. Each affordance has an approval posture you chose.
+3. **The approval and sandbox posture you picked when you started the session.** This determines what Codex can do without asking you.
+4. **The feedback loop.** Tests, type checks, compiler errors, your review. This is how the agent learns it made a mistake in this session.
+
+All four are yours to engineer. None of them are optional. When a harness feels "off", it's almost always because one of these four is underspecified.
+
+---
+
+## 2. Approval modes — pick deliberately, not by default
+
+Codex CLI sessions start with an approval posture. The common modes are:
+
+- **Suggest** — Codex proposes every shell command or file edit and waits for your `y/n`. Maximum control, maximum friction. Good for unfamiliar repos or risky work (migrations, production-adjacent changes).
+- **Auto-edit** — Codex can edit files without asking, but still asks before running shell commands. This is the right default for most feature work in a repo you trust.
+- **Full-auto** (sometimes called "dangerously") — Codex can edit files and run shell commands without asking, inside the configured sandbox. Fastest; appropriate only when the sandbox is actually restrictive and the work is reversible.
+
+**Craft rule:** pick the approval mode *before* you start, based on the blast radius of the work — not during, because the mood of the session shouldn't drive safety posture. Write the mode into your `AGENTS.md` as a note for the next agent.
+
+**Common trap:** raising to full-auto for speed, then forgetting to lower it for the next task. Get in the habit of naming the current mode out loud, the way a pilot names airspeed.
+
+---
+
+## 3. Sandboxing — what "isolated" actually means
+
+Codex runs shell commands inside a sandbox. The sandbox constrains which directories are writable, whether network is allowed, and which binaries are callable. Exact defaults change between versions; what matters is the *posture* you should take:
+
+- **Assume the sandbox is part of the harness.** If it allows network and the task doesn't need network, narrow it. If it allows writes outside the repo and you don't need that, narrow it. Unused affordances are silent drift risks.
+- **Don't ask Codex to weaken its own sandbox.** If a step needs more capability, exit the session, reconfigure, and re-enter with the wider posture explicitly declared. Silent posture drift is the single most common "why did the agent do that" moment.
+- **Document the sandbox assumption in `AGENTS.md`.** One line: "This repo expects Codex with file writes inside `src/` and `tests/` only, no network, shell limited to npm/node." The next agent — human or AI — needs this to know what is load-bearing.
+
+---
+
+## 4. Context window as a resource, not a design aesthetic
+
+The model has a large context window. You do not have a large budget for putting things into it.
+
+**What this means in practice:**
+
+- **Not everything belongs in-context.** `AGENTS.md` should be a map — a short directory of deeper sources — not an encyclopedia. If it's 400 lines, it's a manual, and manuals don't fit in working memory.
+- **Progressive disclosure beats eager loading.** Point at files; let the agent read them when needed. A reference is cheaper than a copy.
+- **Long sessions decay.** After dozens of turns, constraints from the top of the session are in danger of being silently forgotten. Re-surface rules before they matter, not after the agent violates them.
+- **The expensive prompt isn't the one you wrote. It's the one the agent is holding right now.** Every turn, the agent is re-reading the entire session. If there's noise in turn 5, it's still there in turn 50.
+
+**Craft rule:** treat context budget the way a performance engineer treats latency — as a constraint you measure and optimize, not a free resource you assume. The harness is partly a budget discipline.
+
+---
+
+## 5. Long-horizon drift and how to catch it
+
+Long-horizon drift is when the agent gradually stops honoring a constraint you set early in the session, because that constraint has scrolled out of its effective attention. It is the signature failure mode of longer Codex work.
+
+**Symptoms:**
+
+- The agent starts adding `any` types in a repo that explicitly requires strict typing.
+- The agent recreates a utility you told it to import from an existing module.
+- The agent silently stops writing tests for a kind of change it tested earlier.
+- The agent uses the wrong naming convention for the third of three similar files.
+
+**The harness move is not "remind the agent more loudly." It is:**
+
+1. **Put the constraint in a place the agent re-reads.** `AGENTS.md`, a pre-commit check, a test that fails loudly if the convention is broken. Don't rely on the constraint surviving in the scrollback.
+2. **Add a short re-anchor prompt before risky steps.** "Before you implement this, re-read `AGENTS.md` and state the three rules that apply to this change." This is a cheap, reliable drift detector.
+3. **Watch for the moment the agent starts "improvising" around a missing pattern.** That usually means it can't find the pattern any more, not that the pattern doesn't exist.
+
+**Craft rule:** drift is cheaper to prevent with repo artifacts than to correct with more prompting. Every time you correct the same drift twice, encode the constraint.
+
+---
+
+## 6. Before and after — a representative prompt pair
+
+The following is a **representative** comparison, not a live transcript. It is constructed from patterns commonly observed in Harness Lab cohorts. Your actual output will differ; the shape of the difference is what matters.
+
+### The underspecified ask
+
+> Add a dashboard route that shows workshop instances with their current phase and team count. Make it look nice.
+
+What typically happens:
+
+- The agent picks a styling library it guesses is used in the repo (often wrong).
+- It invents a data shape for "instances" instead of reading the existing model.
+- It adds a route at a plausible but not conventional path.
+- It writes no test, because "make it look nice" did not ask for one.
+- You end up with 400 lines of code that almost fit and that you now have to reverse-engineer to reject.
+
+### The specified ask
+
+> **Goal:** Add a dashboard route listing workshop instances with their current phase and team count.
+>
+> **Context:** The instance model lives in `dashboard/lib/workshop-store.ts`. The existing route pattern is in `dashboard/app/workshops/page.tsx`. Styling uses Tailwind + the components under `dashboard/components/ui/`. Do not introduce a new styling library.
+>
+> **Constraints:**
+> - Read-only route. No mutations.
+> - Must work in file-mode storage (local dev) and neon-mode (production).
+> - Follow the existing route naming convention in `dashboard/app/`.
+> - If the instance list is empty, render the existing `EmptyState` component, not a bespoke fallback.
+>
+> **Done when:**
+> - The new page renders the expected columns.
+> - An e2e test in `dashboard/e2e/` loads the page and asserts at least one instance row when the demo data is present.
+> - Running `npm run build` in `dashboard/` produces no new TypeScript errors.
+>
+> Before you implement, read the existing route file and list the three patterns you're going to reuse. Do not start writing until I confirm.
+
+What typically happens:
+
+- The agent stops and reads the referenced files.
+- It lists the patterns it found. You catch any misreading in 30 seconds, before any code.
+- It proposes a plan. You either confirm or redirect — cheap.
+- It implements against the stated done criteria.
+- It runs the test and the build before claiming the work is finished.
+- The work either lands correctly in the first pass, or the feedback loop (test, build, review) catches the gap without you reading every line.
+
+**The difference isn't the wordcount.** It's that the second ask makes the agent's working context match your working context. The harness is what makes that match possible.
+
+---
+
+## 7. A failure-recovery moment
+
+This is the single thing missing from most agent demos: what happens when the agent drifts, and how the harness catches it.
+
+**Scenario (representative, reconstructed from a cohort session):**
+
+A team asked Codex to add a new `facilitator` role to an auth middleware. Codex implemented the change, ran the tests, and reported success. The tests were green. The dashboard worked in dev. Everything looked fine.
+
+The continuation team, the next afternoon, opened the repo and read `AGENTS.md`. One line said: "Any change to auth middleware requires a corresponding update to `docs/adr/` with a rationale." There was no new ADR. The team ran a grep for the new role and found it used only in the middleware — not in the one place in the dashboard that branches on role. The middleware change worked in tests because the test fixtures only exercised the happy path.
+
+**What the harness caught:**
+
+- The ADR rule in `AGENTS.md` caught the missing documentation.
+- The continuation team's "read first, diagnose second" discipline caught the incomplete usage.
+- Neither caught it by "prompting better." Both caught it because the repo carried constraints the morning team had encoded, and the afternoon team had a ritual for reading them.
+
+**The morning team's mistake wasn't prompting. It was trusting green tests as a completion signal in a repo where the real completion criteria lived in a doc they didn't re-read.** The fix is never "write a sharper prompt next time." The fix is either a test that would have failed, a rule that would have blocked merge, or a ritual that would have forced the check. All three are harness moves.
+
+---
+
+## 8. Tool selection — when to reach for what
+
+Harness Lab is agent-agnostic, but in practice you will pick a specific tool for each task. Heuristics:
+
+- **Codex CLI** — best for repo work with fast local iteration, especially when you want a sandbox and shell access. Strong for code generation, refactoring, debugging inside a repo you trust.
+- **Claude Code** — best for long, stateful sessions where you want the model to reason about the full shape of a problem and remember decisions across many turns. Strong for architecture work, careful reviews, careful migrations.
+- **pi** — best for terminal-native work that needs multi-model flexibility. Strong when you want to compare outputs or keep the harness lightweight and scriptable.
+- **Cursor / IDE-native tools** — best for fast edit-edit-edit loops inside a single file or small module, where you want the model's suggestions inline and do not need agentic control flow.
+- **Aider** — best for tightly-scoped edits against a known set of files, with git commits per turn.
+
+**Craft rule:** pick the tool based on the *shape* of the task, not the one you happen to have open. Every tool has an implicit harness; picking the tool is picking the harness.
+
+**None of this means you should switch tools mid-workshop.** Harness Lab cohorts should pick one tool for the day (usually Codex) so the learning accumulates. This section is about Monday morning, not about Saturday's workshop.
+
+---
+
+## 9. How to keep learning after today
+
+The Codex ecosystem ships new capabilities monthly. What's documented here will be partly outdated within the next release cycle. **Do not treat this doc as a frozen reference.** Treat it as a starting harness for your own reading practice:
+
+- Read the official Codex CLI release notes when they ship. The safety and approval posture changes there are the ones that matter most.
+- Subscribe to one practitioner newsletter who uses these tools daily (Simon Willison's blog is a dense source; there are others).
+- When you discover a failure mode in your own work, write it down in your own team's `AGENTS.md` or a runbook. Your team's harness should learn from your team's failures, not just from this doc.
+- Every quarter, re-read your `AGENTS.md` files with a skeptical eye. Delete anything that is no longer load-bearing. Simplicity is part of the harness.
+
+---
+
+## 10. The one-line summary
+
+You cannot prompt your way out of a bad harness. You can, however, engineer a harness that makes prompting mostly unnecessary.
+
+---
+
+## See also
+
+- [`coaching-codex.md`](../materials/coaching-codex.md) — one-page recipe card for the conversational moves that force plan-first work.
+- [`talks/context-is-king.md`](talks/context-is-king.md) — the workshop talk that introduces the method.
+- [`challenge-cards/deck.md`](challenge-cards/deck.md) — small interventions that install the habits during the build phases.
+- [`../workshop-blueprint/day-structure.md`](../workshop-blueprint/day-structure.md) — the full day architecture and the north-star question.