CtrlK
BlogDocsLog inGet started
Tessl Logo

gamussa/coding-policy

Coding policy for Viktor Gamov's AI agents: language-agnostic quality rules, autonomous shipping discipline, and stack defaults for JVM, Swift, TypeScript, and Python

73

Quality

91%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide

SecuritybySnyk

Low

Low-risk findings worth noting

Overview
Quality
Evals
Security
Files

round-flow.mdskills/herdr-teamlead/references/

Round Flow

The shape of one task round, and what the lead does between the steps of skills/herdr-teamlead/SKILL.md. Read this when a round deviates from the happy path.

Compose the Active Team

ResponsibilityRepository writesOutput
developerauthorized source work in its own worktreepushed branch and report
testertest code as a report-directory patchtest plan, patch or verification report
reviewernoneindependent COMMENT review and report
advisor, investigator, architectnonebounded recommendation, diagnosis or design report
releaserelease operations onlyverified release report
judgenonebinding dispute ruling

Activate the responsibilities the next task decision needs. Add specialty requirements through skills/herdr-teamlead/references/specialists.md; a profile on the bench needs no worker or monitoring loop. One worker holds one assignment in a dispatch. Compatible expertise may share a responsibility. Consultations with the same canonical responsibility use separate dispatches, each with its own engagement.

The owner planner applies capability and contribution eligibility before affordable task familiarity and measured headroom. Step 5 in the skill names the executable contract. Use measured costs to improve calibration; preserve the pinned judge and the developer's early-fix reservation.

If no eligible worker fits, record the gap and staff or sequence the work. Never silently drop a required responsibility or let a contributor independently verify its own work. Specialist advice does not replace required reviewer and tester passes.

Two Phases

An implementation task runs through the round twice, and only the second pass gates release. Investigation-only tasks use Step 12's knowledge-deliverable gate; they do not require a pushed branch or release reports.

Phase 1 — pre-development. An architect or advisor supplies the needed design or interaction report. An investigator may first resolve a causal question. Assess each consultation before using its outcome. The tester maps each acceptance criterion to a test, or delivers those tests as a patch (tester Mode A or B). Plan that work with --round tester=test_plan and record its actual contribution through assess-specialist before selecting Phase 2 verifiers. A test patch author is an implementation contributor, even while holding the tester role. The developer implements against both, runs the repo's gates, pushes the branch, and stops without opening a PR.

Work that trips a Team Composition trigger gates here: its deliverable lands before implementation, or the lead records the staffing decision and its reason. Step 5's detect-triggers run decides which of the four fired, and reads that recorded decision; the lead's own reading of the diff does not. The cheap gate is the one worth making mandatory; a reviewer catching the same thing one finding per round is the expensive one.

Untriggered work keeps the old judgement, whatever its size: skip Phase 1 when a design note would say less than the diff. Nothing in Phase 1 is a pass; it is preparation.

Phase 2 — post-push verification (mandatory). The reviewer reviews the pushed branch and posts a COMMENT review (Mode B). The tester runs the gates and the acceptance tests against that same branch (Mode C). Both report against the current tip by SHA.

The release hand-off reads Phase 2 reports and nothing else. A design note is not a review of the code that got written, and a test plan is not a test run.

Phase 2 excludes actual design and implementation contributors from reviewer and tester. The owner applies recorded contribution history; use --exclude for relevant contributions outside it. Assess uncertain prior consultations under skills/herdr-teamlead/references/specialists.md. The reviewer responsibility is verification only; it no longer carries pre-development Mode A. Historical reviewer responsibility remains unknown until its actual contribution is established.

Consultations can enter later when a new question could change implementation or verification. Scope any resulting correction through the accepted behavior and existing correction allowance; a specialist recommendation grants no new implementation authority. A completed consultation returns to the next needed assignment, or closes an investigation-only knowledge deliverable through Step 12.

One Round, End to End

  1. Roster — roster.sh names the live workers. An unnamed pane has no dispatch handle; name it first.
  2. Authority — verify-authority.sh <owner/repo> answers whether the operator owns the repo. Ownership is the namespace, never write permission, and the answer becomes the authority line in every brief.
  3. Measure — teamlead.sh measure reads each worker's own usage numbers. A worker that is working or blocked is skipped with null windows rather than interrupted.
  4. Plan — teamlead.sh plan assigns the requested responsibilities with specialty requirements and explicit contribution exclusions where needed. It contacts nobody; owner state migration may save an older ledger.
  5. Package — Step 6 builds a range-specific VCS artifact for reviewer and tester briefs. Keep the original task base for full reviews and the prior reviewed tip for scoped re-checks; a new tip gets a new package.
  6. Compose — compose-briefs.sh renders the templates from one values file, refusing to write anything when a placeholder is unfilled or a supplied key matches no template. The lead decides the values; the script decides nothing.
  7. Provision — provision-worktree.sh creates every worktree the briefs name, from the shared checkout. A worker never runs git there, so its checkout has to exist before the brief arrives.
  8. Dispatch — teamlead.sh apply uses Step 10's explicit context mode, then sends the assignment prompt. The recovery reference governs developer fix retention and the separate assessed specialist continuation path. It re-reads live status and refuses to type into a busy worker.
  9. Observe — supervision-watch observes every enrolled worker. Verify candidates with wait-report.sh --once, ledger outcomes, and acknowledge handled events under skills/herdr-teamlead/references/supervision.md.
  10. Gate — the lead reads every report in full and decides: another round, or the release hand-off.

Before relying on consultation output, save the report delivery receipt and run assess-specialist under skills/herdr-teamlead/references/specialists.md. Record the accepted outcome in the task ledger and resolve its supervision obligations separately. Keep useful sessions available for likely follow-up, while preserving scoped lessons outside the session. No idle specialist counts as active work.

The lead appends decisions throughout this flow to the persistent task ledger, including before pauses and handoffs. skills/herdr-teamlead/references/task-ledger.md separates dispatch and report observations from assignment acceptance and task completion. All references to the round log here mean that ledger.

Reading a Report

A report is the worker's only channel to the lead. Read all of it, every time — a ## BLOCKED section can sit under a report that otherwise reads as finished.

  • Blocking findings present — take Step 12's bounded fix path under rules/agent-team-operation.md Fix Loops. Keep a stable task identifier and advance its fix counter; changing worker or scope never restarts it. Name the findings and prior report in each brief. Re-check the findings with scoped briefs, then run full verification before release.
  • A ## BLOCKED section — the worker stopped on something it could not decide. Resolve it in the NEXT brief, which reaches it through a fresh dispatch. Never type the answer into the worker that is waiting.
  • wait-report.sh exit 3 — the worker is at an approval or question dialog. Read the pane, relay the dialog text to the operator verbatim, and stop the round for that worker. The lead never answers it — an approval dialog is input, and input to a blocked agent is exactly what Dispatch Safety forbids. The operator answers; the wait resumes once herdr agent get <name> reports a state other than blocked.
  • wait-report.sh --once exit 1 with reason: checkpoint_pending — delivery remains pending. Record the checkpoint, acknowledge its event with a scheduled pending recheck, and resume the fleet watcher. Never send a second copy of the brief on a status hint.
  • wait-report.sh --once exit 1 carrying a stall object — the report is absent, the worker is terminal, and the budget measured from --since is spent. This is terminal for that wait: schedule no further recheck. Read the stall.class, record the user-attention obligation, and take the recovery rules/agent-team-operation.md Stalled Workers names for that class. Never commit the partial work on the strength of the tree building.
  • Persist blocked dialogs, missing reports, and required operator decisions in the attention queue before presenting them. Keep observing unrelated work. A pause or handoff must cover the entire active fleet under the supervision reference; a single worker's blocker does not release those obligations.

Release Gate

For an authorized implementation release, Step 12 requires all four:

  1. The developer's report names the branch and the commit SHA it pushed.
  2. A broad reviewer Mode B report reviews that same SHA and carries no blocking finding.
  3. A broad tester Mode C report verifies that same SHA, with the repo's gates run and every acceptance criterion met.
  4. Nothing has been pushed to the branch after those two reports.

Under a recorded stop remedy, 2 and 3 read against what ships: the excluded defect is a tracked accepted defect and the shipped scope carries no other blocking finding. rules/review-severity.md Judge-Accepted Defect Carve-Out carries the preconditions; every other gate here is unchanged.

A Phase 1 design note or test plan does not satisfy 2 or 3. A report against an older SHA does not either: re-run Phase 2 against the current tip. After scoped re-checks close the findings, re-run Phase 2 with full briefs before handing off to release. Scoped reports alone never satisfy this gate.

Reassess both passes against current contribution history before releasing. A later discovery that a verifier shaped the task's design or implementation invalidates its independence, even when the reviewed SHA is unchanged. Preserve the historical receipt, append the new gate decision and its evidence to the task ledger, and obtain a fresh independent report. Never rewrite the prior review or treat an old approval as authority over newer contribution evidence.

Ancestor-Sensitive Fixtures

Some tools resolve their configuration by walking filesystem ancestors, so a fixture placed under the reports directory is not isolated from the operator's own project: an isolated HOME and the XDG directories are not enough when the lookup starts at the fixture and climbs. One real rehearsal initialized the operator's home-level Tessl project that way, and an unauthenticated run then wrote a rule index missing a private dependency.

The brief names a task-owned fixture root for that work, outside every ancestor that configures the tool, and the worker keeps every report, plan and patch artifact under the reports directory as usual (rules/agent-team-operation.md Writers and Checkouts carries the preconditions).

The root is this assignment's own: created under a name no other assignment uses, never a directory that already exists and never one reached through a symlink. The cleanup at the end removes it, and a reused or linked root would make that cleanup delete somebody else's files.

Inside that root, before any command that writes through the tool:

  1. Resolve the root to its physical path, then walk its ancestors and record which of them configure the tool.
  2. Pre-seed the intended local manifest so the lookup settles inside the fixture rather than above it.
  3. Prove the effective root the tool resolved, and compare it to the fixture. A root anywhere else stops the rehearsal.
  4. Record the state of each user-level file the rehearsal can reach, and read it again afterwards. An unexpected change stops the rehearsal and is reported; restore from that record rather than reinstalling the operator's environment.

Which files those are, and which manifest a given tool reads, belong to the tool's own documentation — not to this reference or a brief.

Blocking Gate

At Step 12, apply skills/herdr-teamlead/references/assignment-reasoning.md to the findings and their proposed corrections. Preserve required judge rulings and operator decisions; scope classification never waives a blocking finding. Read this task's confirmed fix history, name the next fix number, and return to Step 4 with self-contained briefs carrying the findings and prior reports. Preserve the developer for retained fixes; use a fresh context for the fresh-worker stage. Never reset the counter during re-planning. At a contested verdict or a lead override, go to Step 13 first. At an exhausted allowance, record the checkpoint through the owner commands in skills/herdr-teamlead/references/dispatch-recovery.md, report implementation as awaiting_diagnosis, consult the investigator, and go to Step 13 with its assessed report. Use the plan its remedy records for the bounded extra attempts; collect each preceding attempt's actual blocking review before continuing.

Branch-Changing Ruling

At Step 19, the lead never edits the branch itself. At an exhausted allowance, record the checkpoint through the owner commands in skills/herdr-teamlead/references/dispatch-recovery.md and take the diagnosis; the boundary is a diagnostic question, not a budget prompt. The plan its remedy records covers multiple attempts within those bounds; fresh release handoffs do not ask for context-change permission. Report implementation as awaiting_diagnosis while the diagnosis is pending. Record the remedy once and continue within it. Otherwise return to Step 12 carrying ACTION: verbatim as required work. Count that implementation as the next fix, under the same task identifier, and gate the resulting tip again before release.

Why Phase 2 Runs Before the PR

The registry build's PR #27 went through 12 automated review rounds — policy reviewer plus Copilot — because the code reached the bots before the team's own reviewer and tester had seen it. Each round cost a full CI cycle and a context reload on the developer.

A plain branch push does not trigger the policy reviewer. That gap is the opening: the developer pushes the branch and stops. The tester and the reviewer run against the pushed branch, the developer folds their blocking findings in, and only then does the PR open — so the bots review work the team has already agreed on, and the usual outcome is one round instead of twelve.

The hand-off is therefore split around the bots:

  • Developer runs Skill(skill: "release") Steps 1–4 — readiness, PR, version reasoning, and the review request.
  • The bots review once.
  • Developer runs Steps 5–7 — watch the reviews, act on blocking findings, merge and clean up.

The lead releases nothing until it holds a reviewer Mode B report and a tester Mode C report against the SHA the developer pushed. A newer push invalidates both: re-run Phase 2 against the new tip.

Shared-Account Reviews

A Phase 2 review is delivered in the reviewer's report file; there is no PR yet to post it on. When a PR already exists (a fix round after release started), the workers share one GitHub account, and GitHub refuses APPROVE and REQUEST_CHANGES on that account's own PR, so the review is also posted as a COMMENT review with each finding labelled blocking or advisory per rules/review-severity.md. The COMMENT state carries no gate, so the LEAD is the gate: a blocking finding in an internal review sends the round back, whatever GitHub's merge box says.

The Judge

A reserved seat outside ordinary staffing, on the most capable model available. It holds no other responsibility. rules/agent-team-operation.md Judge Seat is the contract; this section is the operational detail for Steps 13–19 of skills/herdr-teamlead/SKILL.md.

It runs in two modes. Adjudication settles a dispute; diagnosis asks why a fix loop is not converging. Dispatch adjudication on exactly one of three triggers:

  • A contested reviewer or tester verdict — one worker's finding, another worker's (or the lead's) disagreement, neither side able to settle it by re-reading the rule.
  • A lead override of a blocking finding — the lead about to waive a finding a worker labelled blocking gets a second, independent read first.
  • A bot finding the team disagrees with — the policy reviewer or Copilot flags something the developer and reviewer both think is wrong.

All three are disputes, and a dispute is settled once. Dispatching an adjudication at every allowance boundary made the seat a per-round toll on the window its developer and reviewers already share; that trigger is gone and stays gone.

An exhausted allowance is a different question, and it gets the second mode — after the investigator. That profile is written for "unclear causality or repeated unsuccessful fixes", it is read-only and bounded, and it is the cheaper seat: it produces the reproduction and causal assessment, and the judge rules on them. A diagnosis without one is refused. The loop that exhausts its budget is rarely short of attempts: a find-rate that holds flat while every round closes its finding is a structural problem, and more rounds reproduce it. Diagnosis asks why the loop is not converging and what has to change. The lead ran the loop and is the wrong diagnostician of its own dispatch pattern, so the read is independent for the same reason a review is.

Its report opens with six lines rather than three:

DIAGNOSIS: <why the loop is not converging, from the evidence>
REMEDY: continue — <rounds, approach unchanged> | restructure — <the change> | stop — <what ships, what is tracked>
BOUND: <developer attempts this remedy allows> — <why that number> | none — for stop
ASSESSMENT: <the investigator report this diagnosis ruled on>
EVIDENCE: <the rounds, findings and diffs the diagnosis rests on>
UNVERIFIED: <anything unconfirmed against the tree, or "none">

A continue or restructure remedy carries the attempt budget in BOUND, which is the number the operator used to supply. It is counted in developer attempts, justified against the cited evidence, and capped: the recording command refuses a bound above its ceiling rather than honouring it. ASSESSMENT names the investigator report the judge ruled on, and the record binds that path the way supervision's enrollment binds the judge's own report. A stop remedy ships what is clean and records the remainder as a tracked accepted defect; that authority is the judge's, stated so a lead does not re-escalate out of caution. Recording it also files a user-attention obligation, so the operator learns of the override they hold without going to look for it.

The diagnosis is re-enterable when its own remedy's bound exhausts with blocking work remaining. A remedy that was independently diagnosed and still did not work is evidence for the next diagnosis, not for the operator, who holds nothing on the second pass they did not hold on the first. Re-entry moves down the ladder continue → restructure → stop, or repeats one rung once against a recorded PROGRESS line: a remedy that produced nothing is never reissued, the ladder never runs backwards, a rung already repeated is spent, and stop is terminal, so a task takes at most five diagnoses. No operator sits in the path of any of them.

rules/agent-team-operation.md Judge Seat carries the contract; the record shapes are the owner's, in skills/herdr-teamlead/references/dispatch-recovery.md.

It is read-only without exception in either mode: no file edit, no mutating git or gh command, no GitHub post, no subagent dispatch.

Adjudicating, it reads both positions and the governing rule, verifies the disputed facts against the tree itself rather than trusting either side's framing, and returns a report opening with three lines — RULING: uphold A | uphold B | amend — <line> | blocked — <question>, ACTION: naming the minimal step, UNVERIFIED: naming anything it could not check — followed by its numbered reasons.

Those three lines belong to adjudication; a diagnosis carries the five above and never a RULING: or an ACTION:.

blocked is the judge declining to rule on a dispute it cannot settle from the tree and the rule text alone. The round stops there and the named question goes to the operator. The lead does not dispatch a second judge and does not rule in its place.

The ruling or remedy binds the round the moment the lead reads it. Only the operator overrides one; record the override and why in the round log. No diagnosis remedy waits on an operator for the task to reach a terminal state. A blocked adjudication is the one ruling that does: it stops the round and sends its named question to the operator, as it always has.

The judge worker is declared in the main config.json and is measured and planned like every other seat, but its seat is pinned rather than ranked: the judge block names the agent, model and effort, the planner never chooses who judges, and the pinned worker never holds another seat.

It runs on the same Claude subscription as the claude worker, so both declare the same window_group and the planner charges a seat's cost against every worker sharing that window. When that window is exhausted the round halts: there is no substitute judge, no fallback to another vendor's flagship, and no degraded ruling.

What the Lead Never Does

  • Edit the shared checkout. The lead reads it and dispatches; workers write.
  • Answer a question by typing into a working worker. Wait for the report.
  • Answer a blocked worker's approval dialog. Relay it to the operator and stop.
  • Create a worktree for a worker after dispatch. Provision before briefing.
  • Write an authority line by hand. It comes from verify-authority.sh.
  • Brief a write action on a repo the operator does not own without their explicit per-repo, per-action permission recorded in the brief.
  • Release on a Phase 1 report. A plan is not a verification.
  • Treat Herdr status as assignment acceptance or task completion.
  • Merge on a worker's behalf. The developer runs the release skill.
  • Act against a judge's ruling, or seat the judge on developer, reviewer, or tester. Only the operator overrides a ruling.

Dispatch Results

Step 10 of skills/herdr-teamlead/SKILL.md follows these outcomes. Before a finish, apply the whole-fleet pause/handoff contract in skills/herdr-teamlead/references/supervision.md.

  • Exit 0 — proceed to Step 11.
  • Busy target — no dispatch occurred. Wait for readiness or replan; stay at this step.
  • Sent but not started — inspect the pane; never re-dispatch on top of the message. Proceed to Step 11 for the roles that started.
  • Retrospective, clear, composer, tier, qualification, or continuity refusal — follow the diagnostic and recorded dispatch outcome. Reconcile uncertainty before retrying; wait for roles whose records confirm dispatch.
  • Unknown refusal — report it verbatim and finish here.
  • --dry-run — inspect the context choice, requested tier, and relaunch argv. It contacts no worker, writes no ledger, and proves no live tier or qualification. Finish here.

Ruling Outcomes

Step 19 follows these branches. Before a finish, preserve any user question and apply the whole-fleet pause/handoff contract in skills/herdr-teamlead/references/supervision.md.

The RULING: line binds the round. Only the operator overrides it.

For an investigation-only task, a non-blocked ruling returns to Step 12's knowledge-deliverable gate with the ruling and any required authorized research. Apply the existing correction allowance and judge rules to remaining findings. Do not enter implementation Phase 2 or Step 20 without implementation/release authorization. A blocked ruling follows the operator-question path below.

  • uphold A / uphold B / amend, ACTION: changing no branch content — record the ruling. Proceed to Step 20 only with Step 12's broad reports against the current tip. Otherwise re-run Phase 2 with full briefs carrying the ruling. Do not re-dispatch the judge for the same settled dispute.
  • uphold A / uphold B / amend, ACTION: changing the branch — apply the round-flow reference's Branch-Changing Ruling contract. Finish here while its required operator decision is pending; otherwise return to Step 12 with ACTION: as the next counted fix.
  • blocked — the judge declined to rule. Stop the round and put its named question to the operator. Do not dispatch a second judge and do not rule in its place. Finish here.
  • REMEDY: continue — record the diagnosis, then return to Step 12 and spend the bounded rounds with the approach unchanged.
  • REMEDY: restructure — record the diagnosis, apply the named structural change to the shape of the work, then return to Step 12 within its bound. The change is the judge's to name and the lead's to carry out.
  • REMEDY: stop — record the diagnosis, release what is clean, and record the remainder as a tracked accepted defect. The remedy carries that authority; do not re-escalate it. Proceed to Step 20 for what ships.
  • A remedy's bound exhausting with blocking work remaining returns to Step 12's exhaustion workflow: record a new checkpoint, consult the investigator again so an assessed consultation follows the latest developer attempt, then take that assessment to Step 13 for the next diagnosis, one rung down the ladder — or at the same rung once, when the spent remedy made progress the new diagnosis records in PROGRESS. The previous assessment never satisfies the new diagnosis. Never re-enter above the last rung, and never repeat a rung twice.

skills

herdr-teamlead

compose-briefs.sh

config.example.json

label-workspaces.sh

provision-worktree.sh

prune-worktrees.sh

resolve-policy-paths.sh

review-package.sh

roster.sh

SKILL.md

start-judge-worker.sh

state-schema.md

teamlead.sh

verify-authority.sh

wait-report.sh

README.md

tile.json