[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M786SAS70MSPYGGFVG7SNH8J":3,"skill-same:01M786SAS70MSPYGGFVG7SNH8J":43},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":15,"categories":16,"keywords":17,"publishedAt":18,"linkOnly":19,"markdownContent":20,"markdownBody":21,"frontMatter":22,"refName":23,"commitSha":24,"skillDirectory":25,"forksCount":13,"repositoryHtmlUrl":26,"repositoryHomepageUrl":6,"ownerHtmlUrl":27,"ownerBlogUrl":6,"ownerDescription":6,"updatedAt":28,"install":29,"resources":42},"01M786SAS70MSPYGGFVG7SNH8J","wayfinder",null,"The orchestration hub for all work. Creates a map of investigation tickets (grilling, research, prototype, task, bug), resolves them one at a time with strict grilling discipline, manages risk posture pauses during implementation, and hands off to to-spec when the map is complete. Use for any work regardless of size — a trivial task has a small map, a large migration has many tickets.","lynchypin\u002Fproject-onboard","lynchypin","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F188926056?v=4","01M786MTZ83KEFY14AD3V2RZ7Y","01M786MTZTPSKAVRY9WFHS792H",0,"MIT","https:\u002F\u002Fgithub.com\u002Flynchypin\u002Fproject-onboard\u002Fblob\u002Fmain\u002Fskills\u002Fwayfinder\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: wayfinder\ndescription: The orchestration hub for all work. Creates a map of investigation tickets (grilling, research, prototype, task, bug), resolves them one at a time with strict grilling discipline, manages risk posture pauses during implementation, and hands off to to-spec when the map is complete. Use for any work regardless of size — a trivial task has a small map, a large migration has many tickets.\n---\n\n# Wayfinder\n\nAll work enters through wayfinder regardless of size, but not all work\nearns a map. Grill first; if the way to the destination is already clear,\ntake the no-fog early exit (Mode 1, step 3) and proceed without one. A\ntrivial task's map might have 1 grilling ticket and 1 task ticket; a large\nmigration might have 20+. The process is consistent; only the map size\nvaries.\n\n## The map\n\nA single document (on the configured tracker) containing:\n\n- **Destination** — one sentence: where are we trying to get.\n- **Decisions so far** — one line per resolved ticket: enough to judge\n  relevance, plus a link to the ticket holding the detail.\n- **Open tickets** — the sharp questions, with blocking relationships.\n- **Not yet specified** — in-scope fog: questions you can tell are coming\n  but cannot yet phrase sharply. Graduates into tickets as the frontier\n  advances.\n- **Out of scope** — work ruled beyond the destination. Closed, and it\n  never graduates.\n\n### The map is an index, not a store\n\nA decision lives in exactly one place — its ticket. The map **gists and\nlinks**; it does not restate. Load the map at low resolution to orient,\nthen zoom into individual tickets on demand.\n\nKeep it this way for a concrete reason: a decision written in two places\nstarts drifting between them the moment one is edited, and a map that\nrestates every decision grows until it no longer fits the context window it\nexists to save.\n\nWhen fog graduates into a ticket, **clear the graduated patch** from *Not\nyet specified* so it lives only as its new ticket.\n\n### Fog or ticket?\n\nThe test is whether you can **state the question precisely now** — not\nwhether you can answer it now.\n\n- **Ticket** when the question is already sharp, even if it is blocked and\n  cannot be acted on yet.\n- **Not yet specified** when you cannot phrase it that sharply. Leave the\n  patch coarse: one patch may graduate into several tickets, or none, once\n  the frontier reaches it.\n\n*Not yet specified* is for work inside the destination. Fog gathers only\n*toward* the destination — work beyond it belongs under *Out of scope*,\nwhere it does not read as takeable frontier.\n\n## Ticket types\n\n| Type | Mode | Purpose |\n|------|------|---------|\n| **Grilling** | HITL | Resolve a decision the user must make. One question at a time. |\n| **Research** | AFK | Investigate primary sources, write findings. Delegates to `research` skill. |\n| **Prototype** | HITL | Throwaway artifact to answer \"how should it look\u002Fbehave?\" |\n| **Task** | AFK\u002FHITL | Config, provisioning, data migration — doesn't need a decision. |\n| **Bug** | AFK\u002FHITL | Diagnose and fix a defect. Delegates to `skills\u002Fdiagnosing-bugs\u002FSKILL.md`. |\n\nHITL = human in the loop. AFK = agent works unattended.\n\n## Grilling discipline\n\nGrilling happens inside tickets typed as \"grilling\". The full grilling\nmethodology lives in `skills\u002Fgrilling\u002FSKILL.md`. Key points:\n\n1. **One question at a time.** Never batch multiple questions.\n2. **Propose before asking.** State a recommended answer; user\n   confirms or corrects. Faster than open-ended questions.\n3. **Facts vs decisions.** If it can be found by reading the codebase,\n   look it up. Only ask about decisions the user must make.\n4. **Challenge every answer.** After each response, explore what it\n   implies, what it breaks, and what happens at the edge. Do not move\n   on until the downstream effects are mapped.\n5. **Spot and reject vagueness.** Push abstract answers (\"flexible\",\n   \"whatever makes sense\") to concrete specifics.\n6. **No question cap.** Could be 3 or 50 — driven by ambiguity. The\n   session ends when you cannot generate another question whose answer\n   would change what gets built.\n7. **Domain modeling inline.** Terms captured to CONTEXT.md as they\n   surface. ADRs written when non-obvious decisions are made.\n8. **Confirmation gate.** Do not proceed from grilling tickets until\n   user confirms shared understanding is reached.\n\n## Mode 1: Chart the map\n\n1. Name the destination in one sentence. It fixes the scope and shapes\n   every ticket, so settle it first.\n2. **Map the frontier** — grill breadth-first, fanning out across the whole\n   space rather than deep on any one thread, to surface the open questions\n   and what is takeable now.\n3. **No-fog early exit.** If that breadth-first pass surfaces no fog — the\n   way to the destination is already clear and the whole journey fits one\n   session — **stop and ask the user how they would like to proceed.** Do\n   not build a map nobody needs. This fires *before* any map exists, and is\n   cheaper than the post-map fork in \"Completion\".\n4. Otherwise create the map: destination filled in, decisions empty, and\n   the fog sketched into **Not yet specified**.\n5. Create the tickets you can specify now, then mark blocking\n   relationships. Everything you cannot yet phrase sharply stays as fog.\n\n## Refer to tickets by name\n\nIn everything the user reads — narration, the map's *Decisions so far* — a\nticket goes by its **name**, not a bare id. `#42, #43, #44` is illegible;\nnames read at a glance. Keep the id or link, but carry it inside the name\nrather than in place of it.\n\n## Mode 2: Work through the map\n\n1. Load the map at low resolution — the index, not every ticket body.\n   Orient to the destination before choosing.\n2. Find the next unblocked ticket. Zoom into related or closed tickets on\n   demand as the work needs them.\n3. Resolve it (grilling → ask user; research → delegate to research\n   skill; prototype → delegate to `skills\u002Fprototype\u002FSKILL.md`;\n   bug → delegate to `skills\u002Fdiagnosing-bugs\u002FSKILL.md`;\n   task → do it).\n4. Record the detail on the **ticket**, and a one-line gist plus link on\n   the map's \"Decisions so far\" — the map indexes, the ticket holds.\n5. Graduate whatever the answer made specifiable into fresh tickets,\n   clearing each graduated patch from **Not yet specified**. Add any newly\n   revealed fog. If the answer shows a ticket sits beyond the destination,\n   rule it out of scope rather than resolving it.\n6. Repeat until all tickets are closed.\n\n## Evidence-bearing STOP gates\n\nA phase does not advance because you assert it is done. It advances when\nyou can show evidence it is done. Before moving a phase forward, PASTE the\nevidence — do not merely claim it:\n\n| Transition | Required evidence (paste it) |\n|------------|------------------------------|\n| Map → direct implement | The resolved decisions list (every grilling ticket closed) + a one-line reason the change fits one context window |\n| Map → to-spec | The resolved decisions list, with every grilling ticket closed |\n| Spec approved | Quote the approved spec section being implemented against |\n| Ticket done | Paste the green test \u002F command output — not \"tests pass\" |\n| Final review done | Paste the combined diff that was reviewed |\n\nAssertion alone (\"tests pass\", \"looks good\", \"that's complete\") is NOT\nsufficient and does not advance the flow. No evidence → the phase stays\nopen. This replaces V1's hard checkpoint state-machine: same resistance\nto premature completion, without a rigid stage enum.\n\n## Tangential work: STOP and classify, never execute\n\nThe single biggest source of drift is doing related-but-off-ticket work\nthe moment it surfaces — the pull to \"just fix this while I'm here.\" That\npull is not a task; it is a signal you have reached the edge of the map.\n\nWhen you notice work that is not part of the current ticket, you MUST:\n\n1. **STOP.** Do not touch it. Do not start \"just a quick fix.\"\n2. **Name it.** State plainly: \"this is off the current ticket.\"\n3. **Classify it** into exactly one of:\n   - **Graduate** → it is in scope and you can state the question sharply:\n     create a new ticket, add it to the map with any blocking\n     relationships.\n   - **Park as fog** → it is in scope but you cannot phrase it sharply yet:\n     log it under **Not yet specified**, never under open tickets. Fog in\n     the ticket list reads as takeable frontier and will be picked up as\n     work before it is ready.\n   - **Rule out of scope** → log it under \"Out of scope\" and close it. An\n     out-of-scope item never silently graduates later without a new\n     decision.\n4. **Return** to the current ticket.\n\nNo classification → no action. This applies during grilling, research,\nand implementation alike. Absorbing the work silently into the current\nticket is the prohibited path.\n\n## Completion → fork: direct-implement or spec-first\n\nWhen all map tickets are resolved:\n\n1. Summarize the complete picture on the map.\n2. Tell the user the map is complete.\n3. **Judge the scope and take the fork.** Ask: does the entire remaining\n   change fit comfortably in a single context window — needing neither\n   multiple sessions nor a written spec to survive a context reset?\n   - **Fits one context** → **direct implement.** Treat the map's\n     accumulated decisions as a single ticket and hand straight to\n     `skills\u002Fimplement\u002FSKILL.md`. Skip to-spec and to-tickets. State the\n     one-line rationale (\"small enough for one context — implementing\n     directly, no separate spec\") so the shortcut is never silent.\n   - **Multiple sessions, or a durable spec would help it survive a\n     reset** → **spec-first.** Hand off to `skills\u002Fto-spec\u002FSKILL.md` to\n     synthesize a spec from the accumulated decisions, then to-tickets.\n\nThis mirrors the fork in the canonical flow: trivial changes go straight\nto implement; larger work earns a spec + tickets so it survives across\ncontext windows.\n\n## Implementation orchestration\n\nAfter to-spec → to-tickets produces the implementation tickets,\nwayfinder orchestrates the implementation phase. **One ticket ≈ one\ncontext window** (see `reference\u002Fcontext-budget.md`): each ticket was\nsized to be implemented and reviewed inside a single fresh context, and\na clean context per ticket is what keeps the agent in the \"smart zone\"\nwhere it doesn't drift or hallucinate.\n\n- For each implementation ticket, invoke `skills\u002Fimplement\u002FSKILL.md`.\n- **After each ticket completes** — commit + tracker update happen inside\n  `implement`; then persist `.agent\u002Fstate.json` (status, next ticket).\n  This makes every ticket boundary a clean resume point: the SessionStart\n  hook re-anchors from state + map, so clearing costs almost nothing.\n- **When recommending a clear, write a handoff first.** `state.json` and the\n  map carry what is settled; they do not carry the live thread — what was\n  tried, what was ruled out, why this approach. Where that reasoning matters\n  to the next ticket, invoke `skills\u002Fhandoff\u002FSKILL.md` and give the user the\n  document path alongside the clear recommendation.\n- **Risk posture behavior:**\n  - `\"standard\"` — may run tickets back-to-back *while the context window\n    is still healthy*. Once the window has grown large or the next ticket\n    is substantial, **recommend the user clear and resume in a fresh\n    session** rather than pushing on in a degrading context: \"Ticket N is\n    done and state is saved — good point to `\u002Fclear` and resume, or I can\n    continue.\" Let the user decide; do not force a reset, and do not\n    silently accumulate every ticket in one window.\n  - `\"careful\"` — after each implement ticket completes, report what\n    was done, whether tests are green, and what the next ticket is.\n    Wait for user OK before continuing; the next ticket resumes in a\n    fresh context.\n- After all tickets pass and per-ticket code-review completes, run the\n  **final combined-diff review** (see below) before anything else.\n- After the final combined-diff review passes:\n  - If `doc_tier` in `.agent\u002Fstate.json` is `\"app\"` or `\"system\"`,\n    **offer an architecture review**: \"Implementation is done. Want me\n    to run an architecture review to check for deepening opportunities?\"\n  - If the user accepts, invoke\n    `skills\u002Fimprove-codebase-architecture\u002FSKILL.md`.\n  - If candidates are found and the user picks one, create new\n    wayfinder tickets for the chosen refactors (these run as a fresh\n    implementation cycle).\n  - If the user declines, skip.\n- Update `.agent\u002Fstate.json` status to `\"done\"`.\n\n## Final combined-diff review\n\nPer-ticket code-review only sees one ticket's diff. It cannot catch drift\nthat spans tickets, or behavior that crept into the combined diff nobody\nasked for. So before completion, run one independent pass over the whole\nchange:\n\n1. **Dispatch a fresh, isolated subagent** — it must NOT be the agent that\n   wrote the code. Delegate to `skills\u002Fcode-review\u002FSKILL.md`.\n2. **Inputs:** the combined diff of all implementation tickets + the\n   approved `docs\u002Fagents\u002Fspec.md`.\n3. **Axes:**\n   - **Correctness** — real defects in the combined change.\n   - **Scope-creep** — behavior present in the diff but absent from the\n     spec. This is the drift the per-ticket pass misses.\n4. **Report** findings most-severe first.\n5. **Gate:** unresolved scope-creep findings block completion. Either the\n   drift is reverted, or it graduates to a new ticket via the\n   STOP-and-classify reflex — it is never waved through silently.\n\n## Bug tickets\n\nWhen resolving a ticket typed as `bug`:\n\n1. Delegate to `skills\u002Fdiagnosing-bugs\u002FSKILL.md`\n2. The skill runs AFK through Phases 1–2 (feedback loop, reproduce)\n3. **Pause at Phase 3** (hypothesise) — present ranked hypotheses to\n   the user for confirmation before testing\n4. Continue AFK through Phases 4–5 (instrument, fix)\n5. Phase 6 post-mortem: if the finding is architectural (\"no good test\n   seam exists\"), create a new wayfinder ticket pointing to\n   `skills\u002Fimprove-codebase-architecture\u002FSKILL.md`\n\n## Prototypes\n\nCreated during \"prototype\" tickets. Delegates to\n`skills\u002Fprototype\u002FSKILL.md`. Two types:\n- **Logic** — terminal app testing state\u002Fedge cases.\n- **UI** — multiple visual variants switchable via URL params.\n\nPrototypes are throwaway. Once the decision is captured on the map,\nthe prototype code is deleted.\n\n## The skill is working if\n\nStated as the behaviour to hit rather than the behaviour to avoid —\nprohibitions make the forbidden move more available in context, not less.\n\n- Implementation begins **once the map is complete** — or once the no-fog\n  early exit has established no map is needed.\n- Every grilling ticket resolves **through the user's own answer**. A HITL\n  ticket resolves only through the live exchange; the agent asking and then\n  answering for the user has, by definition, broken it.\n- Each grilling turn puts **exactly one question** to the user, and waits.\n- New discoveries run the **STOP-and-classify** reflex — graduate, park as\n  fog, or rule out of scope — so every off-ticket pull ends in one of three\n  named places, and the current ticket keeps the scope it started with.\n- Blocking relationships are wired for every ticket, so the frontier is\n  always the open unblocked set.\n- Questions of **fact** get looked up in the codebase; questions of\n  **decision** go to the user. Keeping the split is what stops a grilling\n  agent from racing ahead and answering its own questions.\n","# Wayfinder\n\nAll work enters through wayfinder regardless of size, but not all work\nearns a map. Grill first; if the way to the destination is already clear,\ntake the no-fog early exit (Mode 1, step 3) and proceed without one. A\ntrivial task's map might have 1 grilling ticket and 1 task ticket; a large\nmigration might have 20+. The process is consistent; only the map size\nvaries.\n\n## The map\n\nA single document (on the configured tracker) containing:\n\n- **Destination** — one sentence: where are we trying to get.\n- **Decisions so far** — one line per resolved ticket: enough to judge\n  relevance, plus a link to the ticket holding the detail.\n- **Open tickets** — the sharp questions, with blocking relationships.\n- **Not yet specified** — in-scope fog: questions you can tell are coming\n  but cannot yet phrase sharply. Graduates into tickets as the frontier\n  advances.\n- **Out of scope** — work ruled beyond the destination. Closed, and it\n  never graduates.\n\n### The map is an index, not a store\n\nA decision lives in exactly one place — its ticket. The map **gists and\nlinks**; it does not restate. Load the map at low resolution to orient,\nthen zoom into individual tickets on demand.\n\nKeep it this way for a concrete reason: a decision written in two places\nstarts drifting between them the moment one is edited, and a map that\nrestates every decision grows until it no longer fits the context window it\nexists to save.\n\nWhen fog graduates into a ticket, **clear the graduated patch** from *Not\nyet specified* so it lives only as its new ticket.\n\n### Fog or ticket?\n\nThe test is whether you can **state the question precisely now** — not\nwhether you can answer it now.\n\n- **Ticket** when the question is already sharp, even if it is blocked and\n  cannot be acted on yet.\n- **Not yet specified** when you cannot phrase it that sharply. Leave the\n  patch coarse: one patch may graduate into several tickets, or none, once\n  the frontier reaches it.\n\n*Not yet specified* is for work inside the destination. Fog gathers only\n*toward* the destination — work beyond it belongs under *Out of scope*,\nwhere it does not read as takeable frontier.\n\n## Ticket types\n\n| Type | Mode | Purpose |\n|------|------|---------|\n| **Grilling** | HITL | Resolve a decision the user must make. One question at a time. |\n| **Research** | AFK | Investigate primary sources, write findings. Delegates to `research` skill. |\n| **Prototype** | HITL | Throwaway artifact to answer \"how should it look\u002Fbehave?\" |\n| **Task** | AFK\u002FHITL | Config, provisioning, data migration — doesn't need a decision. |\n| **Bug** | AFK\u002FHITL | Diagnose and fix a defect. Delegates to `skills\u002Fdiagnosing-bugs\u002FSKILL.md`. |\n\nHITL = human in the loop. AFK = agent works unattended.\n\n## Grilling discipline\n\nGrilling happens inside tickets typed as \"grilling\". The full grilling\nmethodology lives in `skills\u002Fgrilling\u002FSKILL.md`. Key points:\n\n1. **One question at a time.** Never batch multiple questions.\n2. **Propose before asking.** State a recommended answer; user\n   confirms or corrects. Faster than open-ended questions.\n3. **Facts vs decisions.** If it can be found by reading the codebase,\n   look it up. Only ask about decisions the user must make.\n4. **Challenge every answer.** After each response, explore what it\n   implies, what it breaks, and what happens at the edge. Do not move\n   on until the downstream effects are mapped.\n5. **Spot and reject vagueness.** Push abstract answers (\"flexible\",\n   \"whatever makes sense\") to concrete specifics.\n6. **No question cap.** Could be 3 or 50 — driven by ambiguity. The\n   session ends when you cannot generate another question whose answer\n   would change what gets built.\n7. **Domain modeling inline.** Terms captured to CONTEXT.md as they\n   surface. ADRs written when non-obvious decisions are made.\n8. **Confirmation gate.** Do not proceed from grilling tickets until\n   user confirms shared understanding is reached.\n\n## Mode 1: Chart the map\n\n1. Name the destination in one sentence. It fixes the scope and shapes\n   every ticket, so settle it first.\n2. **Map the frontier** — grill breadth-first, fanning out across the whole\n   space rather than deep on any one thread, to surface the open questions\n   and what is takeable now.\n3. **No-fog early exit.** If that breadth-first pass surfaces no fog — the\n   way to the destination is already clear and the whole journey fits one\n   session — **stop and ask the user how they would like to proceed.** Do\n   not build a map nobody needs. This fires *before* any map exists, and is\n   cheaper than the post-map fork in \"Completion\".\n4. Otherwise create the map: destination filled in, decisions empty, and\n   the fog sketched into **Not yet specified**.\n5. Create the tickets you can specify now, then mark blocking\n   relationships. Everything you cannot yet phrase sharply stays as fog.\n\n## Refer to tickets by name\n\nIn everything the user reads — narration, the map's *Decisions so far* — a\nticket goes by its **name**, not a bare id. `#42, #43, #44` is illegible;\nnames read at a glance. Keep the id or link, but carry it inside the name\nrather than in place of it.\n\n## Mode 2: Work through the map\n\n1. Load the map at low resolution — the index, not every ticket body.\n   Orient to the destination before choosing.\n2. Find the next unblocked ticket. Zoom into related or closed tickets on\n   demand as the work needs them.\n3. Resolve it (grilling → ask user; research → delegate to research\n   skill; prototype → delegate to `skills\u002Fprototype\u002FSKILL.md`;\n   bug → delegate to `skills\u002Fdiagnosing-bugs\u002FSKILL.md`;\n   task → do it).\n4. Record the detail on the **ticket**, and a one-line gist plus link on\n   the map's \"Decisions so far\" — the map indexes, the ticket holds.\n5. Graduate whatever the answer made specifiable into fresh tickets,\n   clearing each graduated patch from **Not yet specified**. Add any newly\n   revealed fog. If the answer shows a ticket sits beyond the destination,\n   rule it out of scope rather than resolving it.\n6. Repeat until all tickets are closed.\n\n## Evidence-bearing STOP gates\n\nA phase does not advance because you assert it is done. It advances when\nyou can show evidence it is done. Before moving a phase forward, PASTE the\nevidence — do not merely claim it:\n\n| Transition | Required evidence (paste it) |\n|------------|------------------------------|\n| Map → direct implement | The resolved decisions list (every grilling ticket closed) + a one-line reason the change fits one context window |\n| Map → to-spec | The resolved decisions list, with every grilling ticket closed |\n| Spec approved | Quote the approved spec section being implemented against |\n| Ticket done | Paste the green test \u002F command output — not \"tests pass\" |\n| Final review done | Paste the combined diff that was reviewed |\n\nAssertion alone (\"tests pass\", \"looks good\", \"that's complete\") is NOT\nsufficient and does not advance the flow. No evidence → the phase stays\nopen. This replaces V1's hard checkpoint state-machine: same resistance\nto premature completion, without a rigid stage enum.\n\n## Tangential work: STOP and classify, never execute\n\nThe single biggest source of drift is doing related-but-off-ticket work\nthe moment it surfaces — the pull to \"just fix this while I'm here.\" That\npull is not a task; it is a signal you have reached the edge of the map.\n\nWhen you notice work that is not part of the current ticket, you MUST:\n\n1. **STOP.** Do not touch it. Do not start \"just a quick fix.\"\n2. **Name it.** State plainly: \"this is off the current ticket.\"\n3. **Classify it** into exactly one of:\n   - **Graduate** → it is in scope and you can state the question sharply:\n     create a new ticket, add it to the map with any blocking\n     relationships.\n   - **Park as fog** → it is in scope but you cannot phrase it sharply yet:\n     log it under **Not yet specified**, never under open tickets. Fog in\n     the ticket list reads as takeable frontier and will be picked up as\n     work before it is ready.\n   - **Rule out of scope** → log it under \"Out of scope\" and close it. An\n     out-of-scope item never silently graduates later without a new\n     decision.\n4. **Return** to the current ticket.\n\nNo classification → no action. This applies during grilling, research,\nand implementation alike. Absorbing the work silently into the current\nticket is the prohibited path.\n\n## Completion → fork: direct-implement or spec-first\n\nWhen all map tickets are resolved:\n\n1. Summarize the complete picture on the map.\n2. Tell the user the map is complete.\n3. **Judge the scope and take the fork.** Ask: does the entire remaining\n   change fit comfortably in a single context window — needing neither\n   multiple sessions nor a written spec to survive a context reset?\n   - **Fits one context** → **direct implement.** Treat the map's\n     accumulated decisions as a single ticket and hand straight to\n     `skills\u002Fimplement\u002FSKILL.md`. Skip to-spec and to-tickets. State the\n     one-line rationale (\"small enough for one context — implementing\n     directly, no separate spec\") so the shortcut is never silent.\n   - **Multiple sessions, or a durable spec would help it survive a\n     reset** → **spec-first.** Hand off to `skills\u002Fto-spec\u002FSKILL.md` to\n     synthesize a spec from the accumulated decisions, then to-tickets.\n\nThis mirrors the fork in the canonical flow: trivial changes go straight\nto implement; larger work earns a spec + tickets so it survives across\ncontext windows.\n\n## Implementation orchestration\n\nAfter to-spec → to-tickets produces the implementation tickets,\nwayfinder orchestrates the implementation phase. **One ticket ≈ one\ncontext window** (see `reference\u002Fcontext-budget.md`): each ticket was\nsized to be implemented and reviewed inside a single fresh context, and\na clean context per ticket is what keeps the agent in the \"smart zone\"\nwhere it doesn't drift or hallucinate.\n\n- For each implementation ticket, invoke `skills\u002Fimplement\u002FSKILL.md`.\n- **After each ticket completes** — commit + tracker update happen inside\n  `implement`; then persist `.agent\u002Fstate.json` (status, next ticket).\n  This makes every ticket boundary a clean resume point: the SessionStart\n  hook re-anchors from state + map, so clearing costs almost nothing.\n- **When recommending a clear, write a handoff first.** `state.json` and the\n  map carry what is settled; they do not carry the live thread — what was\n  tried, what was ruled out, why this approach. Where that reasoning matters\n  to the next ticket, invoke `skills\u002Fhandoff\u002FSKILL.md` and give the user the\n  document path alongside the clear recommendation.\n- **Risk posture behavior:**\n  - `\"standard\"` — may run tickets back-to-back *while the context window\n    is still healthy*. Once the window has grown large or the next ticket\n    is substantial, **recommend the user clear and resume in a fresh\n    session** rather than pushing on in a degrading context: \"Ticket N is\n    done and state is saved — good point to `\u002Fclear` and resume, or I can\n    continue.\" Let the user decide; do not force a reset, and do not\n    silently accumulate every ticket in one window.\n  - `\"careful\"` — after each implement ticket completes, report what\n    was done, whether tests are green, and what the next ticket is.\n    Wait for user OK before continuing; the next ticket resumes in a\n    fresh context.\n- After all tickets pass and per-ticket code-review completes, run the\n  **final combined-diff review** (see below) before anything else.\n- After the final combined-diff review passes:\n  - If `doc_tier` in `.agent\u002Fstate.json` is `\"app\"` or `\"system\"`,\n    **offer an architecture review**: \"Implementation is done. Want me\n    to run an architecture review to check for deepening opportunities?\"\n  - If the user accepts, invoke\n    `skills\u002Fimprove-codebase-architecture\u002FSKILL.md`.\n  - If candidates are found and the user picks one, create new\n    wayfinder tickets for the chosen refactors (these run as a fresh\n    implementation cycle).\n  - If the user declines, skip.\n- Update `.agent\u002Fstate.json` status to `\"done\"`.\n\n## Final combined-diff review\n\nPer-ticket code-review only sees one ticket's diff. It cannot catch drift\nthat spans tickets, or behavior that crept into the combined diff nobody\nasked for. So before completion, run one independent pass over the whole\nchange:\n\n1. **Dispatch a fresh, isolated subagent** — it must NOT be the agent that\n   wrote the code. Delegate to `skills\u002Fcode-review\u002FSKILL.md`.\n2. **Inputs:** the combined diff of all implementation tickets + the\n   approved `docs\u002Fagents\u002Fspec.md`.\n3. **Axes:**\n   - **Correctness** — real defects in the combined change.\n   - **Scope-creep** — behavior present in the diff but absent from the\n     spec. This is the drift the per-ticket pass misses.\n4. **Report** findings most-severe first.\n5. **Gate:** unresolved scope-creep findings block completion. Either the\n   drift is reverted, or it graduates to a new ticket via the\n   STOP-and-classify reflex — it is never waved through silently.\n\n## Bug tickets\n\nWhen resolving a ticket typed as `bug`:\n\n1. Delegate to `skills\u002Fdiagnosing-bugs\u002FSKILL.md`\n2. The skill runs AFK through Phases 1–2 (feedback loop, reproduce)\n3. **Pause at Phase 3** (hypothesise) — present ranked hypotheses to\n   the user for confirmation before testing\n4. Continue AFK through Phases 4–5 (instrument, fix)\n5. Phase 6 post-mortem: if the finding is architectural (\"no good test\n   seam exists\"), create a new wayfinder ticket pointing to\n   `skills\u002Fimprove-codebase-architecture\u002FSKILL.md`\n\n## Prototypes\n\nCreated during \"prototype\" tickets. Delegates to\n`skills\u002Fprototype\u002FSKILL.md`. Two types:\n- **Logic** — terminal app testing state\u002Fedge cases.\n- **UI** — multiple visual variants switchable via URL params.\n\nPrototypes are throwaway. Once the decision is captured on the map,\nthe prototype code is deleted.\n\n## The skill is working if\n\nStated as the behaviour to hit rather than the behaviour to avoid —\nprohibitions make the forbidden move more available in context, not less.\n\n- Implementation begins **once the map is complete** — or once the no-fog\n  early exit has established no map is needed.\n- Every grilling ticket resolves **through the user's own answer**. A HITL\n  ticket resolves only through the live exchange; the agent asking and then\n  answering for the user has, by definition, broken it.\n- Each grilling turn puts **exactly one question** to the user, and waits.\n- New discoveries run the **STOP-and-classify** reflex — graduate, park as\n  fog, or rule out of scope — so every off-ticket pull ends in one of three\n  named places, and the current ticket keeps the scope it started with.\n- Blocking relationships are wired for every ticket, so the frontier is\n  always the open unblocked set.\n- Questions of **fact** get looked up in the codebase; questions of\n  **decision** go to the user. Keeping the split is what stops a grilling\n  agent from racing ahead and answering its own questions.\n",{"name":5,"description":7},"main","1f0ae11cd93f456cd58d3fd7cd470d1105a00ebf","skills\u002Fwayfinder","https:\u002F\u002Fgithub.com\u002Flynchypin\u002Fproject-onboard","https:\u002F\u002Fgithub.com\u002Flynchypin","2026-08-05 20:49:36",{"source":8,"skillName":5,"commands":30,"prompt":40,"zipAvailable":41},[31,34,37],{"tool":32,"command":33},"npx","npx skills add lynchypin\u002Fproject-onboard --skill wayfinder",{"tool":35,"command":36},"bunx","bunx skills add lynchypin\u002Fproject-onboard --skill wayfinder",{"tool":38,"command":39},"pnpm","pnpm dlx skills add lynchypin\u002Fproject-onboard --skill wayfinder","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Flynchypin\u002Fproject-onboard\u002Fblob\u002Fmain\u002Fskills\u002Fwayfinder\u002FSKILL.md",true,[],[44,51,58,65,72,79],{"skillNo":45,"name":46,"nameZh":6,"description":47,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":48,"categories":49,"keywords":50,"publishedAt":18},"01M786S1VW2DFFP7SGGVYYEN45","to-tickets","Break the spec into small, independently-implementable tracer-bullet vertical slices. No checkpoint gate — hands off directly to wayfinder for implementation orchestration.","https:\u002F\u002Fgithub.com\u002Flynchypin\u002Fproject-onboard\u002Fblob\u002Fmain\u002Fskills\u002Fto-tickets\u002FSKILL.md",[],[],{"skillNo":52,"name":53,"nameZh":6,"description":54,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":55,"categories":56,"keywords":57,"publishedAt":18},"01M786RHCR9MPJP7MKN1T98FAD","to-spec","Synthesize a specification from the completed wayfinder map. Never runs its own interview — it only writes down what wayfinder's grilling tickets already established, at a depth proportional to the project's documentation tier.","https:\u002F\u002Fgithub.com\u002Flynchypin\u002Fproject-onboard\u002Fblob\u002Fmain\u002Fskills\u002Fto-spec\u002FSKILL.md",[],[],{"skillNo":59,"name":60,"nameZh":6,"description":61,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":62,"categories":63,"keywords":64,"publishedAt":18},"01M786R5N3VQCWWDN3CWB4A01S","tdd","Reference material for test-driven development discipline. Red before green, one slice at a time. Use as reference during implementation, not as a step-by-step procedure.","https:\u002F\u002Fgithub.com\u002Flynchypin\u002Fproject-onboard\u002Fblob\u002Fmain\u002Fskills\u002Ftdd\u002FSKILL.md",[],[],{"skillNo":66,"name":67,"nameZh":6,"description":68,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":69,"categories":70,"keywords":71,"publishedAt":18},"01M786QPV9NMXZ4Y6HF1X7MHVF","start-session","Orient a new session in any project directory. Detects empty directory, existing codebase, or resume state, gathers cheap signals, and routes to wayfinder. The only user-invoked entry point.","https:\u002F\u002Fgithub.com\u002Flynchypin\u002Fproject-onboard\u002Fblob\u002Fmain\u002Fskills\u002Fstart-session\u002FSKILL.md",[],[],{"skillNo":73,"name":74,"nameZh":6,"description":75,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":76,"categories":77,"keywords":78,"publishedAt":18},"01M786QAH76026WCTPAVYJAZ79","setup","Configures the issue tracker. Called by start-session during first orient. Model-invoked — runs automatically as part of the flow, not user-triggered.","https:\u002F\u002Fgithub.com\u002Flynchypin\u002Fproject-onboard\u002Fblob\u002Fmain\u002Fskills\u002Fsetup\u002FSKILL.md",[],[],{"skillNo":80,"name":81,"nameZh":6,"description":82,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":83,"categories":84,"keywords":85,"publishedAt":18},"01M786Q0PSCKJAJ4YSGHXBXV8V","scope-lock","Produces a short context manifest of which files and directories are in-bounds for the current task on a medium or large existing codebase. Keeps subsequent reading inside that boundary unless explicitly widened. Prevents context bloat.","https:\u002F\u002Fgithub.com\u002Flynchypin\u002Fproject-onboard\u002Fblob\u002Fmain\u002Fskills\u002Fscope-lock\u002FSKILL.md",[],[]]