agentleFS
sponge-b0bsponge-b0b/Polaris/.agents/skills/implement-ticket/SKILL.md

implement-ticket

Skill4 starsChanged 3 months agoApache-2.0View on GitHub

Implement one exact deterministic ticket contract, run applicable technical checks, obtain fresh independent closure proof, and close only the exact certified candidate.

18 other projects use a version of this.

Adopted by

Most recently taken up by

Used with

Projects with it also use thermo-nuclear-code-quality-review, implement-spec and 1 more.

Other files people use alongside this one.

Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
  • jaydubya818Skill

    thermo-nuclear-code-quality-review

    Use when the user requests an intentionally strict maintainability review.

    Also in 3 projects that use this

  • AWSSkill

    implement-spec

    Implement the result of /to-spec and /to-tickets in code.

    Also in 3 projects that use this

  • GoogleSkill

    to-spec

    Turn the current conversation into a spec and publish it to the project issue tracker — no interview, just synthesis of what you've already discussed.

    Also in 3 projects that use this

The file

Show its rules and text

We split this skill into the rules it gives an agent

17 rules. 0 are common, 8 are distinctive, and 9 are very specific to this project.

  • 0

    Common

    Most of the 17 edited versions we read keep these.

  • 8

    Distinctive

    This author's own choices. Worth a look.

  • 9

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Before executing this skill, read all of these files in full:

    line 12
  2. Very specific:

    Only an explicit resume $implement-ticket - <Title> (<URL>) invocation may consume an old journal.

    line 41
  3. Very specific:

    All retained requirements remain mandatory unless explicitly overridden here, BOUNDED_EXECUTION.md, or POST_ROUTE_RETURN.md on their narrow …

    line 46
  4. Distinctive:

    Before ordinary Ticket-baseline handling, inspect durable recovery state exactly as required by the shared contract.

    line 59
  5. Distinctive:

    If the exact current default-branch validator cannot be executed and proven, fail closed.

    line 83
  6. Distinctive:

    When native order places Source transitively upstream of Destination:

    line 91
  7. Very specific:

    after exact certification/push/readback/closure, update the same recovery record to source-complete with the exact Source completion commit …

    line 100
  8. Very specific:

    after exact Destination closure, populate Destination baseline/completion commit and terminalize the same record as source-reconciled with …

    line 105
  9. Distinctive:

    When native order places Destination transitively upstream of Source, preserve the historical prerequisite-extract lifecycle from the …

    line 110
  10. Very specific:

    after Source closure, populate its exact completion commit and terminalize as source-reconciled.

    line 117
  11. Very specific:

    When the recovery record has Source disposition: retire, the protected Source donor is recovery evidence, not an executable participant.

    line 127
  12. Very specific:

    before the Source is durably closed-not-planned, do not begin Destination implementation.

    line 132
  13. Distinctive:

    if the caller checkout is dirty, ahead, diverged, stale, or otherwise fails the ordinary branch/Execution-Anchor guards, preserve it and …

    line 142
  14. Distinctive:

    before transplant, recompute the protected donor fingerprint and re-attribute every donor item against the current Destination contract.

    line 145
  15. Very specific:

    after exact Destination certification, commit, push, readback, and closure, populate the exact Destination baseline/completion commit and …

    line 152
  16. Very specific:

    Every terminal Human Handoff from $implement-ticket remains bound to the shared gate, including verifier cell-routing-defect -> …

    line 213
  17. Distinctive:

    Do not stop after the first actionable ticket.

    line 242

“Common” is measured: kept by at least 6 in 10 of the 17 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Implement Ticket

Authoritative procedure composition

Before executing this skill, read all of these files in full:

  1. .agents/skills/implement-ticket/ROUTING_CONTROL_PLANE.md
  2. .agents/skills/to-tickets/TICKET_RESLICE_RECOVERY.md
  3. .agents/skills/WORKFLOW_INVOCATION_JOURNAL.md
  4. .agents/skills/implement-ticket/WORKFLOW_INVOCATION_JOURNAL.md
  5. .agents/skills/implement-ticket/BOUNDED_EXECUTION.md

ROUTING_CONTROL_PLANE.md preserves the previously authoritative $implement-ticket entrypoint and recursively composes PROCEDURE.md, EXECUTION_ANCHOR.md, and CONTROL_PLANE.md, including the protected workflow/control-plane boundary, durable routing defects, deterministic ticket contract, Attention, closure, and shared Human Handoff rules. CONTROL_PLANE.md additionally composes POST_ROUTE_RETURN.md for the narrow existing-authority return-to-implementation boundary. TICKET_RESLICE_RECOVERY.md is the canonical durable contract for every ticket-reslice-recovery:v1 record and overrides the shorter recovery summary retained in PROCEDURE.md. WORKFLOW_INVOCATION_JOURNAL.md at the skills root owns the shared WIJ storage/schema/non-authority contract; this skill's WIJ document owns only $implement-ticket integration and resume classification. BOUNDED_EXECUTION.md owns pre-closure concern admission, post-verifier-failure bounded triage, mandatory durable stage checkpoints, and the narrow Durable Continuation terminal. On those subjects it overrides retained procedure and skill-local WIJ language, including the retained three-terminal-condition limit. On the post-route return boundary, POST_ROUTE_RETURN.md as composed by CONTROL_PLANE.md wins. On any other conflict, this file wins. Failure to load any retained execution/recovery/bounded-execution/route-return authority file is a Hard Blocker.

WIJ is a best-effort execution optimization, never a lifecycle prerequisite. Journal failure, loss, corruption, ambiguity, or incompatibility must degrade to the ordinary current $implement-ticket reconstruction and must not create a new blocker or termination state. Only an explicit resume $implement-ticket - <Title> (<URL>) invocation may consume an old journal; an ordinary invocation starts fresh. The workflow-durable implement-ticket-continuation:v1 comment defined by BOUNDED_EXECUTION.md is distinct from WIJ and is the only source that can authorize the narrow Durable Continuation terminal.

All retained requirements remain mandatory unless explicitly overridden here, BOUNDED_EXECUTION.md, or POST_ROUTE_RETURN.md on their narrow owned subjects.

Execution-anchor invariant

Retain the accepted distinction between the immutable historical Ticket baseline and the current attempt's Execution Anchor. Do not require the current shared branch to equal that historical SHA. Before a fresh attempt, fetch the actual remote ticket branch and, only while clean, fast-forward only when the local branch is strictly behind and is an ancestor. The retained Execution Anchor and Attention-gate rules remain mandatory.

Live-WIP recovery consumption boundary

Before ordinary Ticket-baseline handling, inspect durable recovery state exactly as required by the shared contract. Any current recovery record that names this ticket as Source or Destination must be read back and machine-validated before substantive mutation:

python -m scripts.workflow_reslice_recovery validate \
  --body <recovery-comment-body-file>

Require TICKET RESLICE RECOVERY: VALID. Malformed, duplicate, contradictory, ambiguous, or native-order-inconsistent recovery state is a Hard Blocker.

Recovery validator execution source

scripts.workflow_reslice_recovery is workflow control-plane runtime, not product candidate content. Resolve and fetch the current remote default-branch tip before validating a recovery record. When the caller branch does not contain that helper, or its helper blob is not exactly the current default-branch blob, execute the validator from a clean detached checkout/worktree at the exact fetched default-branch tip and pass the recovery-comment body by path. Do not copy, cherry-pick, merge, or synchronize the validator or its workflow tests into the active Spec branch merely to validate recovery state.

Recovery validation is read-only with respect to both the caller candidate and the protected donor. It must not checkout, reset, stash, clean, commit, or otherwise mutate either worktree. If the exact current default-branch validator cannot be executed and proven, fail closed; do not fall back to a stale candidate-branch helper or manually recreate the validation.

Native dependencies determine actionability. Recovery metadata never makes a blocked ticket actionable and never changes its deterministic ticket contract.

Source-first recovery

When native order places Source transitively upstream of Destination:

  • Source executes first from an isolated worktree at the exact current remote shared-branch tip while preserving its pinned historical Ticket baseline;
  • the isolated worktree receives a fresh Execution Anchor and a fresh pre-mutation Attention receipt bound to the current reconciled Ticket Contract Hash;
  • an old Attention receipt from the protected donor is recovery evidence only;
  • transplant only Source-owned donor items. shared dependency items remain Source-owned and become ordinary committed upstream behavior for the later Destination;
  • after exact certification/push/readback/closure, update the same recovery record to source-complete with the exact Source completion commit and machine-validate/read back it;
  • intermediate native tickets execute normally;
  • Destination executes only when its blockers close, from an isolated worktree at current remote tip; if its Ticket baseline is Pending, pin that isolated HEAD and read it back;
  • after exact Destination closure, populate Destination baseline/completion commit and terminalize the same record as source-reconciled with both completion commits.

Destination-first recovery

When native order places Destination transitively upstream of Source, preserve the historical prerequisite-extract lifecycle from the shared contract:

  • Destination executes first in an isolated worktree and, after closure, updates the record to prerequisite-complete with exact Destination baseline/completion commit;
  • Source later executes from a new isolated worktree at current remote tip containing the prerequisite completion, re-attributing donor WIP against current contracts;
  • after Source closure, populate its exact completion commit and terminalize as source-reconciled.

For either order, the protected donor checkout remains immutable. Recompute its strong fingerprint before every Human Handoff/ordinary return and before retiring any isolated participant worktree. The second participating closer terminalizes the same managed record; do not create a replacement record.

Retired-source destination recovery

When the recovery record has Source disposition: retire, the protected Source donor is recovery evidence, not an executable participant. Preserve the shared contract's retirement, attribution, fingerprint, transplant, and terminalization requirements, with this execution- location correction:

  • before the Source is durably closed-not-planned, do not begin Destination implementation;
  • once the record is source-retired with Source retirement: closed-not-planned, the record's Recovery mode: isolated-worktree means that the protected donor checkout remains isolated and immutable; it does not require the Destination implementation to run in a separate worktree;
  • the Destination follows the ordinary $implement-ticket shared-Spec-branch rules. When the caller checkout is clean, is on the exact declared Destination branch, tracks its exact remote branch, and can synchronize to the fetched remote tip by fast-forward only, execute there normally. Do not create a separate implementation worktree merely because the retired-source recovery record exists;
  • if the caller checkout is dirty, ahead, diverged, stale, or otherwise fails the ordinary branch/Execution-Anchor guards, preserve it and fail closed or use an isolated worktree only where the ordinary caller-worktree safety rules independently require one;
  • before transplant, recompute the protected donor fingerprint and re-attribute every donor item against the current Destination contract. Copy only still-Destination-owned bytes into the Destination candidate; never mutate, commit, reset, clean, or checkout-overwrite the protected donor;
  • capture a fresh Destination Execution Anchor and fresh contract-bound pre-mutation Attention receipt in the actual Destination execution checkout. A donor Attention receipt remains recovery evidence only;
  • after exact Destination certification, commit, push, readback, and closure, populate the exact Destination baseline/completion commit and terminalize the same managed record as source-reconciled.

This retired-source correction does not weaken execute-source recovery: source-first and destination-first records continue to use their required isolated participant worktrees.

Retained protected control-plane boundary

An ordinary $implement-ticket invocation still has zero authority to alter repository workflow/control-plane or semantic-authority files. There is deliberately no $implement-ticket bypass flag.

For $implement-ticket candidate admission, the older direct scripts.workflow_candidate_guard check / restore-protected invocation is superseded by the policy-sync-aware wrapper required by CONTROL_PLANE.md:

.agents/skills/implement-ticket/scripts/candidate_guard.py

The canonical root guard still owns the protected-path definitions consumed by that wrapper. This override changes attribution of already-committed authorized policy sync only; it does not permit candidate-local edits to protected paths.

Independent verifier result-finalization boundary

The retained dispatcher-only rule is strengthened as follows. After receiving one exact PASS/FAIL payload from the fresh $verify-ticket-closure context, do not yet admit or route that verdict. First compute SHA-256 over the exact returned bytes and send that same verifier context one fixed result-finalization request containing only that digest and the instruction to execute its authoritative result-finalization contract.

During this handshake the parent remains dispatcher-only. It must not debate a finding, provide new evidence, suggest that the verifier is wrong, request a different semantic interpretation, or perform its own semantic appeal. The same verifier must return exactly one of the authoritative forms defined by $verify-ticket-closure:

  • TICKET CLOSURE RESULT: CONFIRMED bound to the exact result SHA-256 and exact ticket / contract / Attention / Execution Anchor / candidate identities; or
  • TICKET CLOSURE: INVALID with Invalid reason: self-contradictory-verdict, bound to the exact retracted result SHA-256 and the same identities.

Mechanically reject a confirmation/retraction whose digest or identities do not match the exact dispatched attempt.

A confirmed result proceeds through the retained PASS/FAIL admission path unchanged, except that a confirmed FAIL must pass the bounded verifier-failure triage in BOUNDED_EXECUTION.md before any corrective candidate mutation. A self-retracted result is neither PASS nor FAIL: preserve the original PASS/FAIL bytes as retracted evidence in the v3 attempt history, record the attempt as invalid, keep the candidate immutable, retire that verifier context, and dispatch one new fresh verifier if candidate/contract/coverage/Attention bindings remain exact.

$implement-ticket may not create an INVALID payload, infer self-contradiction on the verifier's behalf, or semantically overrule a confirmed PASS/FAIL. Free-form follow-up with a verifier after PASS/FAIL is prohibited; the one fixed result-finalization handshake is the only same-context post-result interaction.

Retained Human Handoff boundary

Every terminal Human Handoff from $implement-ticket remains bound to the shared gate, including verifier cell-routing-defect -> $to-tickets, verifier planning-cell-gap -> $to-specs, and contract-authority-gap routing. A recommendation, prose instruction, or bare downstream skill command is not a Human Handoff.

The exact downstream artifact rules remain:

  • $to-tickets receives the exact current Parent Spec issue.
  • $to-specs receives the exact current planning source for the Parent Spec.
  • A current Spec issue is a durability source, not a substitute planning source.
  • $implement-ticket receives each exact current actionable ticket issue.
  • $verify-spec receives the exact current Parent Spec issue.

Finalize terminal handoffs only through the shared runtime resolver:

python .agents/skills/shared_handoff_runtime.py finalize \
  --manifest <handoff-manifest.json> \
  --output <handoff-output.txt> \
  --repo-root .

The resolver may use candidate-local scripts.workflow_handoff only when it satisfies the current versioned shared runtime contract. Otherwise it fetches current origin/main, materializes only the canonical shared handoff runtime closure in a temporary location, and generates/verifies there without mutating the candidate branch.

Require WORKFLOW HANDOFF: VALID and reproduce the generated block byte-for-byte. For a normal post-closure frontier, evaluate the complete current actionable ticket frontier; Do not stop after the first actionable ticket. Emit one manifest handoff entry for each current actionable ticket. Do not emit a bare $to-specs, $to-tickets, $architecture-remediation, $implement-ticket, or $verify-spec line as a terminal Human Handoff.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.