Repository navigation
Releases
v2.9.1
Patch Changes
- 37b9a31: Glossary entries and ADRs now ship with the code that makes them true, never on the default branch ahead of it.
to-ticketsmakes them an Epic's first Sub-issue, which blocks every other one, so they are the first merge into the integration branch. For a single issue,hardenowns the files and carries their text, so they arrive in that issue's pull request.
v2.9.0
Minor Changes
- 27003e7:
hardennow drops a Test scenario an existing test already covers, merges scenarios that differ only by input, and names the test a new scenario replaces. - 27003e7:
reviewnow accepts a criterion thathardenmarked as already covered by an existing test, and for abreak not runmark it runs the deliberate break and reports the criterion unproven unless the named test turns red. The review panel's testing reviewer accepts both forms of the mark and leaves the break toreview. - 27003e7:
tddnow declines a test another test already covers, and refuses tests of constants, message wording, framework behaviour and trivial code. - 27003e7: Added
/test-audit, which reads a test suite, reports the tests another test already covers, and files the removals you pick as issues for triage. - 27003e7: The review panel's
testingPersona now reports a test the pull request adds that another test already covers, and names the test to keep. - 27003e7: Triage now proves an issue filed by
/test-auditby breaking the code in a scratch copy and running the Keeper, and closes it aswontfixwhen the Keeper stays green.
Patch Changes
- a0b1f1f: In
auto-grill, every veto line now opens with a handle (V1,V2and so on, numbered across the stop, lines that depart from common practice first), so "strike V3" is a complete reply. In the Claude desktop app, each drawing is sent right after its own question, and a button sends the letter answer (1B) on its own line, so two clicks no longer run together.
v2.8.0
Minor Changes
- e9ef54f:
/chief-of-staffnow keeps a Wiki: a folder of dated, linked markdown pages above your repos, with one page per goal and an index. It recommends an existing Wiki and uses it only on your yes, cites what a repo records instead of copying it, and supersedes a changed claim instead of rewriting it. - e9ef54f:
/chief-of-staffnow sends work out instead of doing it: it starts a Build session for an Epic or One-off and a triage session for a filed issue, answers a waiting session as your Proxy when a source settles the question, relays the ones that touch the gate, runs research in background subagents, and asks you to type the user-invoked steps. - e9ef54f: Added
/chief-of-staff, an experimental session role that acts as your Proxy for the goals you give it: it settles what the record settles, brings you Escalations, Challenges and a veto list, and asks before merging to a protected branch, releasing or closing an Epic. - 4ab3f54: A
grillingRound can now be answered with letters: every option sits on its own line asA:,B:and so on, the recommended one first, with aRests on:line naming what the recommendation assumes, and a reply such as1A 2Borallanswers the whole Round, with a caveat after any letter. Each Round after the first reads back the replies that changed the plan, says how many choices are settled and which are held back, and lists the facts it looked up with their sources. The closing restatement marks each line with its source (you, the record, or the agent's inference) and puts the inferred lines last as the ones to check. Inauto-grill, the veto list stands in for the list of looked-up facts.
v2.7.0
Minor Changes
- 67611e4: A session can now triage an issue with the new
triagingskill: tell it "use triaging for #N" and it triages N exactly as/triage Ndoes, so an agent such as a Chief of Staff session can send out one triage session per filed issue. The triage steps, references and credits move fromtriagetotriaging, and every message in whichtriagingwaits on you opens with#<N> waits on you:./triagestill shows the waiting issues when you name none, and now asks whether to triage here or in a sub-session where sessions can be started.
v2.6.0
Minor Changes
- 9bdb63d:
/build <N>in the desktop app now asks one question first: run the build here, or in a sub-session. It recommends the sub-session when your session has already started other sessions or sits outside N's checkout. A sub-session starts in the checkout of N's repository (found one folder down when you are above it, or asked when there are none or several) and runsbuildingthere, and an answer you give in your own session to its question is passed on to it in your words. In a terminal nothing changes:/buildbuilds here without asking. The choice lives in a new shared skill,delegating. - 9bdb63d: A session can now start a build with the new
buildingskill: tell it "use building for #N" and it builds N exactly as/build Ndoes, so an agent such as a Chief of Staff session can start a build without anyone typing/build. The build's steps, references and credits move frombuildtobuilding;/buildstill works as before and now hands the issue tobuilding. - 9bdb63d: Every worker is now a subagent. The
desktopandbgDrivers are gone, and so is--driver:spp dispatch <issue>takes no flags and writes the brief without startingclaude --bg,spp planno longer records or renders a Driver, andspp nextno longer puts adriveron itsdispatchactions. Passing--driverto either command is a usage error.buildingno longer asks which Driver to use at the first Wave; its approval question now comes first in its message, so a parent session sees it. An Epic whose plan comment still records a Driver builds as before, with subagent workers. - 9bdb63d: A build resumed after its session died now asks about the work its lost workers left behind.
spp next <epic> --resumereturns each Sub-issue that is claimed, with no open or merged pull request and no assignee, as ahand-backsayingworker lost, andbuildingpasses--resumeon the firstspp nextof each run, so you choose whether to hand the issue back or start a worker on it again.spp next,spp wait,spp wave planand the Epic board no longer read theclaude agentssession list:spp waitno longer printst-<N> donelines, and the board lists only its own session's subagent workers, so itsattachbutton is gone.
Patch Changes
- 9bdb63d:
spp nextandspp wave plantake--running <tickets>: the tickets the build started a worker on, until that worker reports back, count as in flight, so neither command dispatches them again, and a ready ticket that overlaps one is held. Without the flag nothing is held that way.buildingkeeps that list and passes it on everyspp next, and starts each worker's Agent description witht-<N>, so the Epic board links the worker to its row. The board's worker entries drop theirkindfield, and a failing agent list read shows asagent list: <message>. - 9bdb63d:
spp nextandspp wave planshare one parser and one error message for the flag.buildingpasses each running ticket in the form thedispatchaction named it, and resumes a worker by the agent id its Agent call returned. - 9bdb63d: A resumed build no longer starts a fresh worker on a branch another worker left. When a fix round, a rebase or a settled design needs a worker that died with an earlier session,
buildingasks you whether to hand the issue back or take the pull request over.spp brief's first step again only cutssdd/<N>-<slug>from the base and confirms it, and a restarted lost worker starts on that fresh branch, so a branch the lost worker pushed is yours to delete or the issue goes back.spp next --runningnow holds a running ticket in any state, landed included: neither it nor its fix pull requests are routed again while its worker works, andmain-gatewaits for it.spp wave plan --runningholds each one until it lands, with the reasonrunning, its worker has not reported back.buildingkeeps each worker's agent id for the rest of the session and puts an issue back on the running list when it resumes its worker. - 9bdb63d:
building: each answer to the dead-worker question now leaves a markspp nextreads, so the samefixormergestops coming back. Taking a pull request over puts its issue on the running list until you say you are done; handing back a fix pull request closes it with its open findings, so the main gate can run.spp brief's first step now stops BLOCKED when origin already has the ticket's branch or cannot be reached, and the brief tells the worker never to force-push. - 9bdb63d: Every question
buildingstops on now opens with one line,#<N> waits on you: <question>: the first Wave's approval, a lost worker, a red Verify after a merge, a design to settle and an issue it cannot build.delegatingrelays your answer from a parent session to any of them, not only the Wave approval./buildon a Sub-issue or a raw issue now stops with the reason before it asks where to run, and starts no sub-session. Restarting a lost worker removes its local worktree and branch once you agree.
v2.5.0
Minor Changes
- c415a22:
spp lint pr <pr>checks a pull request body for the review gate's sections: theClosesorRefsline,## Ticketand## Laws, a Verify line on the head commit,Verify block corrected:andInterfaces changed:. Theimplement-issueworker runs it at step 8 and fixes the body before it reports, so a missing section no longer costs a fix round.
Patch Changes
-
c2cb705:
spp status,spp next,spp retro, the epic board andspp mergenow count a closing reference only when a line of the pull request body opens with a closing keyword (spaces or tabs may come before it), such asCloses #2orFixes: acme/app#2. A sentence that says "this also fixes #1's edge case" no longer links the pull request to #1, and a reference followed by an apostrophe is never a closing reference, even at the start of a line. Before this, a Fix pull request that mentioned another ticket in prose could be listed as that ticket's pull request with its verdict, andspp mergecould refuse it withgate not passedfor the wrong ticket. -
54a8c7d:
spp findings mergecounts one defect once when review-panel Personas title it differently: findings on one file within five lines merge whatever their titles, the body lists each other title as anAlso reported as:line, and aticket-mandated:,Candidate ADR:,Check after deploy:orTerminologytitle merges only with its own tag. With--fix-diff, a merged finding counts as on a changed line when any finding it absorbed was. -
ae19d85:
spp retrono longer undercounts review rounds.spp wave <epic> paneltakes a new--pr <pull request URL>and keeps each pull request's rounds apart, so a Multi-repo epic's second pull request no longer overwrites the first's round 1; the wave comment lists each round asPanel round 1 on owner/repo#N, andspp retroadds up every pull request's rounds and findings. The review-panel skill now passes--pr. A call with no--prwrites what it wrote before.spp retroalso counts the failing gates on Fix pull requests, found asspp statusfinds them. They show as a newfixPrRoundstally besiderounds, which still counts the Sub-issues' pull requests, and their findings count inbyCauseand under the Sub-issue each fix refs. With no integration branch on the plan comment,fixPrRoundsreads"not found". -
9435c98: The review skill now starts its fresh reviewer with the pull request and nothing else, so an orchestrator no longer adds its own criteria, round histories or setup notes to the reviewer's prompt.
v2.4.0
Minor Changes
-
42bec5a: Pending Skill feedback is now offered at checkpoints: at the end of each Wave and of each run in
/build, in/close-epicstep 5, and at the start of a new session through aSessionStarthook. Each runs the newspp feedback checkpoint, which lists the candidates and the steps to file, skip or keep each one, and prints nothing underCI, outside an Attended session, with the Self-improvement mode Off or on a worker'ssdd/branch. Under Submission mode Review the person answers each preview; under Auto spp files them./close-epicno longer writes a drafts comment (<!-- sdd-feedback-draft -->): unattended runs collect nothing, so there is nothing to draft. Its[skill]learnings are recorded as Signals and filed throughspp feedback file. -
42bec5a: A new model-invoked
feedbackskill records a wish or complaint about spp or one of its skills as awishSignal throughspp signal, using the person's own words as evidence. It starts only from the person's own chat message, never from an issue, file or tool output, and never for a wish about their own project. -
42bec5a:
spp feedback file <fingerprint>files a pending candidate on the one grouped Skill feedback issue for its skill step and category on syn54x/skills-plus-plus, creating it with theskill-feedbacklabel when none is open, and rewrites your own tally comment there (Signals and runs per plugin version). Under your Submission mode Review it prints what it would post and exits 3, and files only when run again with--confirm <hash>for that text; under Auto it files at once, except in a private repo. With--epic, the strictest Self-improvement mode of the Epic's repos applies. The outbound check runs on everything it posts, and nothing is filed underCIor with the mode Off.spp feedback skip <fingerprint>drops a candidate for good, andspp feedback pendingnow leaves out skipped candidates and filed ones until a newer Signal arrives.spp feedback submission [Auto|Review]reads or sets your Submission mode, kept inconfig.jsonin spp's config directory. -
42bec5a: Every Marker comment spp writes now records who made it. Its data trailer's
bynames the skill (to-tickets for the plan comment, build for the wave comment, review for the verdict, implement-issue for the progress comment) and the plugin version, and the progress comment'sverifyPinBynames the Verify pin's maker (harden, or triage when it re-pins).spp plan,spp wave <epic> add|pr|panel,spp progress,spp verdict,spp hand-backandspp lint ticket --pintake--by "<skill> <step>"to name another skill and the step. Comments written before this change carry no maker. -
42bec5a:
spp feedback checknow scans every skill-feedback body for secrets (GitHub, GitLab, AWS, Google Cloud, Slack, Anthropic, OpenAI, npm and Stripe tokens, JWTs and private keys, with patterns taken from gitleaks' default rules) and blocks a body holding one in every Self-improvement mode. Each hit prints the secret as[REDACTED]. Under Full Detail the six other rules now print aswarning:lines and exit 0, so you see what a body says about your project without being stopped by it; under Anonymous they still block. -
42bec5a:
spp signal <category> --by "<skill> [step]" --evidence "<text>"records a Signal, a correction charged to the skill step and plugin version that made the corrected work, in a per-repo Signal log beside the notes log. It records only in an Attended session, never underCI, and never when the repo's Self-improvement mode is Off.spp feedback pendinglists the recorded Signals grouped by skill step and category.The Routing block's
Self-improvement mode:line (Off,AnonymousorFull Detail) replacesUpstream feedback:. An old line still works:offreads as Off,countsanddetailedas Anonymous, andattributedas Full Detail.spp preflight --jsonnow printsselfImprovementin place ofupstreamFeedback. -
42bec5a: A new
spp gate <pr>command reads a pull request's slice gate the wayspp mergedoes: the verdict comment from a trusted author, judged against the pull request's current head. It printsgate-pass,gate-failorpr-openwith the head and each result, and--jsonadds the verdict and the maker that wrote it. A skill that asks whether the gate passed a pull request now runs one command instead of decoding the verdict trailer.
Patch Changes
-
42bec5a: The attended decision
spp dispatchpasses to a worker now lives in one file per ticket, so dispatches started together no longer overwrite each other's. A worker sees it from its first command (ansdd/<N>-orsdd-<N>branch, or aclaude --bg --worktree sdd-<N>folder before it cuts its branch) and honours it only while the ticket's issue is open. Only the issue closing ends it, since the same worker is resumed after its slice merges and after a needs-info hand-back: a worker that finds the issue closed deletes it.spp note --auto, the Stop hook and the epic board now recognise every worker shape too. Claude Code's attended variable is now named inhooks/attended.json, which the npm package also ships, rather than in spp's source. -
cddc6ca:
auto-grilldraws each choice so it reads on a narrow screen. Every Escalation and Challenge now opens with the question in one plain sentence, then stacks its options underA:andB:labels within 72 columns, with the Best fit marked and only names you already know. Before this, options sat side by side in columns over 100 characters wide, which wrapped in the desktop app and buried the deciding difference. In the Claude desktop app, where an inline visual tool is available, each choice is drawn as rendered cards, grids or timelines, with a button per option that sends your answer. -
42bec5a: Four commands now record the corrections they already see as Signals, charged to the skill step that made the corrected artifact: a failed
spp verdict(fix-round), eachspp hand-backroute (brief-gap,missed-edge,human-needed), eachspp plan --add-deviation(contract-edited), and the ready gate'spin-matchinspp lint ticketwhen a trusted person edited the body (contract-edited).spp feedback pendinglists them. The same Signal is never recorded twice, and nothing is recorded underCI, outside an Attended session or with the Self-improvement mode Off. Exit codes, stdout and GitHub writes do not change.Every Marker comment write that creates the comment or hands it to another skill step now counts one run in a runs log beside the Signal log, the denominator for signals per run.
spp signalunderCIin a checkout with no GitHub origin now printsnot recorded: CIand exits 0 instead of failing, and a Signal that cannot be written is reported rather than thrown, sospp findings mergekeeps its exit code. -
42bec5a: A Signal about another repo's issue now keeps that repo's name (a new optional
aboutfield, also onspp feedback pending --json), andspp feedback filetreats those repos as it treats an Epic's: Submission mode Auto falls back to Review when one is private, and the outbound check guards their names. A run is recorded once:spp findings merge --fix-diffnames its run by the patch's content (fix diff <hash>, afterpull request <N>when the patch sits in the panel'sspp-panel-<N>folder), so merging the same round again adds none, and a fresh panel run whose round numbers start again at 1 still counts its rounds. When a skill step has no recorded runs, the grouped issue's title and body and the tally comment say its runs are not recorded instead of reporting 0 (the tally's trailer storesruns: null), and thefeedbackskill charges a wish to the skill alone, so it is counted over that skill's runs.spp feedback checkpointchecks for a worker before anything else, so a worker's session start asks gh nothing. The spp config file, which holds the tally secret, is left at mode 0600 on every write. -
42bec5a:
spp feedback filenow adds your tally only to a grouped issue that you or an owner, member or collaborator ofsyn54x/skills-plus-plusopened, so a stranger's issue carrying the fingerprint can no longer collect it; the Review preview names the issue's author. You keep one tally comment per repo you file from, keyed by a hash of the repo under a secret kept in your spp config, so the key names no repo and filing from a second repo no longer overwrites the first one's counts.spp feedback filenow refuses outside an Attended session and in a worker, and the checkpoint recognises a worker the same waysppdoes everywhere else. -
42bec5a:
spp findings mergetakes a new--fix-diff <file>flag. From round 2 the review panel passes it the patch of the fixes made since the previous round. A merged finding that sits on a line the patch added or changed, and that norecheck-finding was merged into, is recorded as afix-regressionSignal against the fix round step ofimplement-issue, and stderr says whether each one was recorded. Without the flag, and in stdout and--jsonwith it, the merge behaves as before. -
42bec5a:
spp findings merge --fix-diffnow records one run ofimplement-issuestepfix roundfor every merge, whether or not a finding regressed. Before this, nothing recorded fix rounds as runs, so the groupedfix-regressionissue reported "over 0 recorded runs" and the per-version rate could not be read. -
e2e0b67: A worker resumed for a fix round now plans every finding before its first edit, following
implement-issue's newFIX-ROUND.md. For each finding it writes down the root cause, every caller and test the change reaches, any overlap with another finding's fix, and the test that guards against a regression. Then it adds a## Fix plansection to the pull request. Before this, a worker fixed each finding at the line the finding named, and the fix often br...
v2.3.0
Minor Changes
- 2d5b628: The epic board now lists an Epic's fix pull requests in a
Fixesgroup after the Waves: one row per open or merged fix with afix openorfix mergedchip, its number, the Sub-issue itsRefsline names (orunattributed) and aPRbutton that copies its link. While the Epic has any fix, the board header and the status line end with the open count, as in12/12 merged · 3 fixes open. A failing pull request read keeps the ticket rows and shows its error. A board stored by an earlier version is dropped and redrawn on the next refresh. - 2d5b628:
spp mergenow merges a Fix pull request: one that closes no Sub-issue of its Epic butRefsa Sub-issue, with a trusted passing verdict on its head, the Epic's integration branch as base and passing checks. It writes nothing to the wave comment, and a pull request that closes no issue and refs no Sub-issue is refused. - 2d5b628:
spp nextnow routes an Epic's Fix pull requests the way it routes a slice:gatewhen its checks pass and its head has no verdict,fixonce after a failed head (naming the failing runs when checks are the cause),hand-backnaming the pull request after a second failed head,mergeon a passing verdict, andwaitwhile its checks are pending or its Sub-issue's worker is running. An unattributed fix is never gated or merged: it is reported underwait, orstuckwhen nothing else is in flight, with the way out (addRefs #<Sub-issue>to its body, or close it).main-gatenow prints only once no fix is open. The action shape is unchanged. - 2d5b628: The
reviewskill now gates a Fix pull request: one with no closing line that opensRefs #Nfor a Sub-issue and targets its Epic's integration branch gets both verdicts, the Sub-issue's Verify block, and a verdict marker recorded withspp verdict --ticket N.build's Main gate followsspp nextfor fixes (gate,fix,mergeorhand-back) andspp mergemerges them, so the orchestrator no longer merges a fix by hand. - 2d5b628:
spp statusnow lists an Epic's Fix pull requests: trusted pull requests into its integration branch that close no Sub-issue, open or merged (ADR 0011). The text view prints them under aFixesheading after the Layers, one line each with its state, its number and the Sub-issue itsRefsline names orunattributed, and the header line ends with the open count, as in(3 fixes open); with no fixes it prints neither.--jsoncarries the same list asfixeson the Epic state, each withticketset to the Sub-issue's URL or null.spp waitexits with a change line when a fix appears, merges or goes (#77 OPEN -> MERGED), gets a new head after a fix round or gets the gate's verdict (#77 OPEN -> OPEN (new head)). The snapshot makes one more GraphQL request per repo of the Epic (its own, each Sub-issue's and each the plan names) for the list, only when the plan comment records an integration branch; a repo that is not there or not readable is warned about on stderr once and loses only its fixes, while any other failure fails the command.snapshot.tsexportspullRequestsArgs(snapshot),fixRepos,withPullRequestsandreadPullRequestsfor the epic board. - 3c14c8d:
/to-ticketsnow checks the set for values worked out twice. When one Sub-issue works a value out from another module's output (a client inferring each row's state from a summary the server returns), it asks whether the module that holds the inputs should produce the value itself. If so, the producer gains a Produces line or a Sub-issue of its own before anything is released, and the consumer reads the value. A derivation you meant to keep is yours to accept.
Patch Changes
- 8856ea4:
spp dispatch <issue> --driver subagentnow refuses a folder that is not inside a git repository, before it reads GitHub. Thespp:workeragent cuts its worktree from the session's folder, so every worker started from such a folder failed withCannot create agent worktree: not in a git repository. The refusal says to run/buildfrom inside the checkout of the repository that owns the issue, or to choose another Driver. ThebganddesktopDrivers are unchanged. - 2d5b628: A Fix pull request in
EpicState.fixesnow carries itsgateon the current head, computed once, andspp nextreads it. The way out for an unattributed fix namesRefs #<Sub-issue>only, sinceCloseson a closed Sub-issue makes the pull request that ticket's own. - f197712: The stop hook,
spp verify --check, now ends its refusal on aStopevent with one more sentence: if this is a worker's worktree and you are not that worker, cd back to your own checkout and stop again. The hook judges the tree the session stands in and cannot tell whose work is there. An orchestrator whose shell was left inside a worker's worktree was therefore refused at the end of every turn until the worker committed, and one run polled through 18 fix rounds that way.buildon Claude Code now tells the orchestrator to read a worker's worktree withgit -C <path>and to stay out of it. ASubagentStoprefusal and a run by hand print what they printed before, and which stops are refused has not changed.
v2.2.0
Minor Changes
- 5a6a9ac: New skill
discuss: type/discuss <idea>, or say you want to brainstorm or kick an idea around, and the agent talks a vague idea over with you before it is a plan, then names where it should go next. It is experimental, so it may change or be removed. - 1ca3abf:
/epic-boardwith no argument now picks the Epic itself: the parent of ansdd/<N>-*branch, the Epic whose plan records afeat/<slug>branch, or the one Epic in flight in the repo. With several in flight it lists them; with none it says so.
v2.1.0
Minor Changes
- b402fe2: In the Claude desktop app,
/buildnow asks which Driver to use when you approve an Epic's first Wave, since desktop linked sessions start each later Wave only on a message from you.spp plan <epic> --driver desktop|bg|subagentrecords the choice andspp nextreports it ondispatch, so a resumed run reuses it.
Patch Changes
- 5cc096f: Credit each skill's sources: a skill that drew on an incumbent now lists them under
metadata.creditsin its frontmatter, with aCREDITS.mdbeside it saying which part came from where.