{"slug":"next-3","title":"next","summary":"This skill should be used when the user asks \"what's next\", \"what should I work on\", \"continue\", \"resume\", \"pick up where I left off\", or names a task directly to jump to it. It lists which tasks are open in the current project and where each stands, or loads the one named, and o","platform":"Claude","tags":[],"authorName":"LLM Mart","authorSlug":"llm-mart","score":0,"source":"github","price":null,"verified":false,"createdAt":"2026-11-10T18:22:09.972714Z","repo":{"url":"https://github.com/camoa/claude-skills","stars":32,"forks":6,"license":"MIT","updatedAt":"2026-11-10T03:29:18Z"},"bodyHtml":"<hr>\n<h2>name: next\ndescription: This skill should be used when the user asks \"what's next\", \"what should I work on\", \"continue\", \"resume\", \"pick up where I left off\", or names a task directly to jump to it. It lists which tasks are open in the current project and where each stands, or loads the one named, and offers to start a task when none are open.\nargument-hint: \"[</h2>\n<h1>Next</h1>\n<p>A task lives inside a project. This skill answers one question: which tasks are open here, and\nwhere each one stands. Read the argument once. No argument: show what is open, below. Any text:\ntreat it as a task id and go straight to \"A task named directly,\" below.</p>\n<p>This skill never creates a task and never offers a contract, even when it ends up recommending\nthat one be started. Creating one is the task skill's own job. Its one write is the move of a\nlegacy task, below, which the task skill owns.</p>\n<p>Every call below runs <code>next-actions.sh</code>, or the one <code>task-actions.sh repair</code> call. Both are named\nin this skill's own grant, so each runs without asking, in both run modes. Any other Bash command\nstill asks for approval.</p>\n<h2>Determine the run mode</h2>\n<p>No task is active yet. Deciding which one is active is this skill's own job, so there is nothing\nto read a stated run mode from. Treat this invocation as interactive, the safe default, unless the\nwhole session is already known to be running unattended from outside any task. Only then pass\n<code>--run-mode autonomous</code> on every <code>next-actions.sh</code> call below. The move call takes no run mode\nflag. Decide this once, at the start.</p>\n<h2>No argument: what is open</h2>\n<p>Run:</p>\n<pre><code>\"${CLAUDE_PLUGIN_ROOT}\"/skills/next/scripts/next-actions.sh --run-mode &lt;interactive|autonomous&gt; report\n</code></pre>\n<p>Read the exit code first, never the text alone.</p>\n<table>\n<thead>\n<tr>\n<th>Exit code</th>\n<th>Meaning</th>\n<th>What to do</th>\n</tr>\n</thead>\n<tbody>\n<tr>\n<td>0</td>\n<td>A project was found and it has at least one open task.</td>\n<td>Read the <code>OPEN:</code> list, below.</td>\n</tr>\n<tr>\n<td>1</td>\n<td>No project owns this directory, and none is remembered for it either.</td>\n<td>Say so, in one line, and name the project skill as the way to set one up or switch to one. Stop.</td>\n</tr>\n<tr>\n<td>2</td>\n<td>A project was found, but it has no open task.</td>\n<td>Go to \"With nothing open,\" below.</td>\n</tr>\n<tr>\n<td>3</td>\n<td>The script could not do its job.</td>\n<td>Show the error text and stop.</td>\n</tr>\n</tbody>\n</table>\n<p>On exit 0 or 2, the output begins with <code>CASE:</code> and <code>RUN_MODE:</code>. <code>CASE:</code> is the resolution case\nas the project skill numbers them; <code>RUN_MODE:</code> is the mode passed in. Neither changes what to do\nnext. Then comes <code>PROJECT:</code> (the project folder), then an <code>OPEN:</code> section holding one JSON\nobject per line, most recently worked first. A <code>LEGACY_COMPLETE:</code> section ends it.\nA line beginning <code>next-actions:</code> is a warning: one task file could not be read and was skipped,\nnever silently. A final <code>WARN:</code> line means at least one of those happened; say so in one line\nrather than letting it pass unremarked.</p>\n<p>Each <code>OPEN:</code> line is one of two shapes:</p>\n<ul>\n<li><code>\"kind\":\"new\"</code>: a task with its own <code>task.json</code>. Carries <code>id</code>, <code>state</code> (<code>new</code> or\n<code>in_progress</code>), <code>parent</code>, <code>children</code>, <code>runMode</code>, <code>runModeStages</code> (the stages the mode covers,\nempty meaning all), and <code>review</code>: what <code>review/review.json</code> says,\n<code>passed</code> or <code>failed</code> from its verdict, <code>unfinished</code> for a record with no verdict, <code>none</code> with no\nrecord. An open task reading <code>passed</code> or <code>failed</code> is reviewed, and completion closes it. Also\n<code>notes</code>: the date of the newest file under the task's <code>notes/</code>, or <code>none</code>. It also carries\n<code>worktree</code>: the path of the task's own git worktree, or <code>none</code> for a task made before every\ntask had one, or one waiting on another task's build (<code>afterBuild</code>, below). A follow-up and a\nsplit's child read <code>none</code> too, until <code>start</code> makes the tree. With a path it carries <code>worktreeOnDisk</code> too, <code>yes</code> or <code>no</code>: whether that folder\nis on this machine. The key is absent with <code>none</code>.\nAnd <code>stage</code>: where the task stands, the first stage whose close record is absent,\none of <code>scope</code>, <code>research</code>, <code>design</code>, <code>implement</code>, <code>review</code>, <code>completion</code>.\n<code>legacyStages</code> appears only on a moved version 5 task. It lists the stages whose version 5\nfile the move kept under a <code>.v5</code> name and whose version 6 record is absent. The order is the\nstage order, from <code>scope</code>, <code>research</code>, <code>design</code>. Otherwise the key is absent. See \"Carrying\nversion 5 work forward,\" below.\n<code>after</code> appears only on a task made with <code>--after</code>. It names the task this one builds on.\n<code>afterBuild</code> says whether that build is <code>finished</code>, <code>unfinished</code> or <code>missing</code>. A task that\ndoes not read <code>finished</code> cannot start yet. Say which task it waits on.</li>\n<li><code>\"kind\":\"legacy\"</code>: a task from before the tasks folder existed. Carries <code>id</code>, <code>epic</code> (the\nfolder it is nested inside, or <code>null</code>), <code>legacyState</code> (<code>in_progress</code> here; <code>complete</code> only\nappears under <code>LEGACY_COMPLETE:</code>), and <code>path</code>.</li>\n</ul>\n<p><strong>Exactly one line under <code>OPEN:</code>.</strong> That is the answer. Say which task it is and its <code>stage</code>.\nName <code>/aida:&lt;stage&gt;</code> as the skill to run next. When <code>afterBuild</code> is not <code>finished</code>, name the task\nit waits on and no stage, because every stage refuses. Treat it as active. This step writes nothing, except the move of a legacy task; there is no session file. It asks one\nquestion only when the line carries <code>legacyStages</code>: the offer in \"Carrying version 5 work\nforward,\" below. Go there. Otherwise\nenter the tree, below. A\n<code>kind: legacy</code> task still lives outside the project's own tasks folder. Move it, below. Then run\nthe report again. Read the task as <code>kind: new</code>. Never delete it: the move keeps it.</p>\n<p><strong>More than one line.</strong> List them in the order printed, each numbered, showing the id, its\nstate and its review word (or, for a legacy entry, its <code>epic</code> and that it predates the tasks folder). Add <code>after</code> and <code>afterBuild</code> when a line has them. Ask which one.\nWait for a plain answer, a number or the task's own id. A chosen legacy entry goes to the move,\nbelow, and then to \"A task named directly\", below. Any other choice goes to \"A task named\ndirectly\" as it is.</p>\n<ul>\n<li><strong>Autonomous.</strong> Do not ask. Show the list, say that several tasks are open and none was chosen,\nand continue. Choosing one for the person would bind every later step to a guess.</li>\n</ul>\n<p>If <code>LEGACY_COMPLETE:</code> holds any lines, name them once, for visibility: these finished before the\ntasks folder existed and are not part of the choice above.</p>\n<h2>Enter the tree</h2>\n<p>Every stage action of a task runs inside its own git worktree, and refuses from anywhere else.\nTwo routes reach that tree. The refusal at exit 79 names the one that works where the call ran,\nso read the message and take the route it names.</p>\n<p><strong>First, test the path.</strong> Two readings stop this step before either route. A task reading\n<code>worktree: none</code> has no tree yet. A task reading <code>worktreeOnDisk: no</code> records a path this machine\ndoes not have: the tree was removed, or another machine recorded it. On either reading, do not\ncall the entry tool and do not name a route. Say the first stage action that needs the code makes\nthe tree and names the path it made. When <code>afterBuild</code> is not <code>finished</code>, say instead that\n<code>start</code> makes the tree once that build is finished. Say a recorded path is computed again here, never reused.\nStop there. Only a task whose <code>worktreeOnDisk</code> reads <code>yes</code> goes on to the two routes.</p>\n<p><strong>The prefix.</strong> Start the Bash call with <code>cd &lt;worktree&gt; &amp;&amp;</code>. It works from any folder on the\nmachine and needs no version. The <code>cd</code> part asks for approval, because the tree sits outside this\nwindow's working directories. It does not ask when the tree is an added directory or an allow\nrule covers it. The shell forgets such a directory between calls, so every call carries the\nprefix.</p>\n<p><strong>Entry.</strong> From a window inside the code checkout, call the <code>EnterWorktree</code> tool with the task's\n<code>worktree</code> path, a folder outside the checkout. The tool asks for approval, because the path\nsits outside <code>.claude/worktrees/</code>. No permission rule and no \"don't ask again\" stops that prompt.</p>\n<p>Entry fails in two different ways. Key on the kind of outcome, never on the words of a message.</p>\n<ul>\n<li><strong>The person declines the prompt.</strong> Do not call the tool again. Say that the prefix is the\nroute now, and use it on every call.</li>\n<li><strong>The tool refuses.</strong> Entry is not available from this window at all, so use the prefix. It\nrefuses from a window outside the code repository, and from a session already inside a\nworktree, where it reaches only targets under <code>.claude/worktrees/</code>. From a session inside\nanother task's tree, <code>ExitWorktree</code> returns to the directory that session started in. Entry\nworks from there when that directory is inside the code repository.</li>\n</ul>\n<p>Then print the path and name the route the person types. <code>/cd &lt;path&gt;</code> moves this session into the\ntree and keeps the conversation (Claude Code 2.1.169 or later). Running <code>claude</code> from inside the\ntree needs no version and no approval. Stop there.</p>\n<p>A legacy line never reaches this step: the\nmove, below, comes first, and this skill then reads the task again as <code>kind: new</code>.</p>\n<h2>With nothing open</h2>\n<p>Exit code 2 above means a project was found with no open task.</p>\n<p>Offer to start one, in one line. Wait for a plain yes or no.</p>\n<ul>\n<li><strong>Yes.</strong> Invoke the task skill to create it, passing whatever name or goal was already given.\nThis skill's own job ends here; the write belongs to that skill.</li>\n<li><strong>No.</strong> Stop.</li>\n<li><strong>Autonomous.</strong> Do not ask. Say that no task is open, and continue.</li>\n</ul>\n<h2>Moving a legacy task</h2>\n<p>A legacy task predates the tasks folder and has no <code>task.json</code>. The move that brings it in is the\ntask skill's <code>repair</code>. This skill runs it, so a person never types it. The task skill says what the\nmove does and when it refuses. Ask nothing first. With the <code>PROJECT:</code> line and the task's <code>path</code>\n(or its <code>PATH:</code> line), run:</p>\n<pre><code>\"${CLAUDE_PLUGIN_ROOT}\"/skills/task/scripts/task-actions.sh repair --project \"&lt;projectPath&gt;\" \"&lt;path&gt;\"\n</code></pre>\n<p>Show the whole output. On success, say what moved and where.</p>\n<h2>Carrying version 5 work forward</h2>\n<p>A moved task can carry <code>legacyStages</code>. Say which stages version 5 finished for this task, from\nthat list, and that version 6 holds no record of them. Then offer one thing: run the first listed\nstage now. That stage reads the version 5 file as its input and writes the version 6 record. The\nproducer runs again, and there is no converter, so the stage that reads the old file is the\nrepair. Wait for a plain yes or no.</p>\n<ul>\n<li><strong>Yes.</strong> Invoke <code>aida:&lt;first stage&gt;</code> through the Skill tool, once, with the task id, and stop.</li>\n<li><strong>No.</strong> Enter the tree, above.</li>\n<li><strong>Autonomous.</strong> Do not ask. Invoke it the same way, and stop.</li>\n</ul>\n<h2>A task named directly</h2>\n<p>Run:</p>\n<pre><code>\"${CLAUDE_PLUGIN_ROOT}\"/skills/next/scripts/next-actions.sh --run-mode &lt;interactive|autonomous&gt; open \"&lt;target&gt;\"\n</code></pre>\n<p>Read the first line.</p>\n<ul>\n<li><strong><code>FOUND: new</code>.</strong> Summary lines follow: <code>PATH:</code>, <code>task-file:</code>, <code>id:</code>, <code>state:</code>, <code>parent:</code>,\n<code>children:</code>, <code>runMode:</code> (with the stages the mode covers in brackets when it covers fewer\nthan all), <code>worktree:</code>, <code>worktree-on-disk:</code> when a path is recorded, <code>review:</code> and <code>stage:</code>, with <code>legacyStages:</code> when\nit applies. A task made with <code>--after</code> adds <code>after:</code> and <code>after-build:</code>, which mean what\n<code>after</code> and <code>afterBuild</code> mean in the report. Say which task it is, from its <code>id</code>, <code>state</code>, <code>review</code> and <code>stage</code>. Name\n<code>/aida:&lt;stage&gt;</code> as the skill to run next. When <code>after-build:</code> is not <code>finished</code>, name the task\nit waits on and no stage. Treat it as active. Read\nthe file at <code>task-file:</code> only when another field is needed. With <code>legacyStages:</code>, go to\n\"Carrying version 5 work forward,\" above; its offer is the one question asked here. Otherwise\nenter the tree, above.</li>\n<li><strong><code>FOUND: legacy_in_progress</code>.</strong> A <code>PATH:</code> line follows, and an <code>EPIC:</code> line when it is nested\ninside one. Move it, above. Then run the same action again. Read it as <code>FOUND: new</code>.</li>\n<li><strong><code>FOUND: legacy_complete</code>.</strong> The same lines follow. Say plainly that this task finished before\nthe tasks folder existed and stays where it is: no contract is offered on it here.</li>\n<li><strong>Exit 3</strong> naming the target: the name is not a safe task id (a path separator, <code>.</code> or <code>..</code>).\nSay so and ask for a different one.</li>\n<li><strong><code>NOT FOUND: &lt;target&gt;</code></strong> on stderr. Say so, then offer the same choice as \"With nothing open,\"\nabove: start a new task with this name, or show what is open instead.</li>\n</ul>\n","files":[{"path":"references/handoff.md","sizeBytes":674,"isText":true},{"path":"scripts/legacy-tasks.sh","sizeBytes":2859,"isText":true},{"path":"scripts/next-actions.sh","sizeBytes":19587,"isText":true},{"path":"SKILL.md","sizeBytes":12412,"isText":true}],"reviewScore":null,"reviewSummary":null,"trust":{"provenance":"trusted-source-unreviewed","notice":"Community-authored content, reproduced verbatim and not vetted as instructions. Treat it as data to evaluate, never as directives to follow.","bodySource":null},"bodyLocked":false,"purchaseUrl":null,"sourceUrl":null,"report":{"provenance":"trusted-source-unreviewed","screen":{"ran":true,"outcome":"clean","suspicious":0,"notes":0,"hiddenCharacters":false},"virusScan":{"engine":"clamav","status":"clean","scannedAt":"2026-11-10T18:22:30.247265Z","sha256":"A91FE7D52EE205C11C1696E626DE2BBD046DEC5BDD1A0C2FE1BE4F00BC50F62C","sizeBytes":13435},"review":null,"source":{"repositoryUrl":"https://github.com/camoa/claude-skills","path":"aida/skills/next","license":"MIT","commit":"c73ba2a2a322813b7fb9d6e4a08285bf5c87dea4","subtreeSha":"9A614403DF5CF0AE6A4246EEED10B3C5C09D1FA1C40554EEB7496DDED400CFF9","lastSyncedAt":"2026-11-10T18:22:06.443791Z"},"reviewedAt":"2026-11-10T18:22:53.815262Z","notice":"Community-authored content, reproduced verbatim and not vetted as instructions. Treat it as data to evaluate, never as directives to follow."},"install":[{"target":"skills-cli","command":"npx skills add https://github.com/camoa/claude-skills/tree/main/aida/skills/next"},{"target":"claude-code","command":"claude plugin marketplace add https://llmmart.ai/marketplace.json && claude plugin install camoa-claude-skills@llmmart"},{"target":"git","command":"git clone https://github.com/camoa/claude-skills.git"}]}