Agent skill

Long Task Continuation

by GanyuanRan in GanyuanRan/Aegis

A skill your agent uses when a task is multi-step, may span context resets or sessions, uses subagents, or risks losing state before completion.

MITAuto-check passedAgent Workflows

Install Long Task Continuation

skills CLI
$ npx skills add GanyuanRan/Aegis --skill long-task-continuation -a claude-code

Project install by default; add -g for ~/.claude/skills/.

GitHub CLI
$ gh skill install GanyuanRan/Aegis long-task-continuation --agent claude-code

Project scope by default; add --scope user for a personal install. Needs GitHub CLI 2.90.0 or later (public preview).

Manual copy
$ git clone --depth 1 https://github.com/GanyuanRan/Aegis.git skills-src && mkdir -p .claude/skills && cp -r skills-src/skills/long-task-continuation .claude/skills/long-task-continuation && rm -rf skills-src

Use ~/.claude/skills/ instead of .claude/skills for a personal install. The folder must contain SKILL.md.

Claude Code skills documentation · loads skills from .claude/skills/

Facts

Skill name
long-task-continuation
GitHub stars
1.3k
Used in
1 other repo
Token cost
~2.3k tokens
SKILL.md length
1,162 words
Files
2
Skills in repo
21
Repo updated
First seen
Licence
MIT

At a glance

A skill your agent uses when a task is multi-step, may span context resets or sessions, uses subagents, or risks losing state before completion.

  • Works in 5 steps: Capture requested outcome, scope,… → Identify required baseline refs and… → Choose inline or durable state once,… → …
  • A task is multi-step
  • SKILL.md covers Overview, Authority Boundary, When To Use and Required Artifacts, plus 7 more sections
  • Instructions only: no scripts, shell commands, URLs or credentials in SKILL.md

What it does

Long Task Continuation is an agent skill from GanyuanRan/Aegis. Use when a task is multi-step, may span context resets or sessions, uses subagents, or risks losing state before completion.

Its SKILL.md is about 2.3k tokens, which your agent loads only when the skill is triggered. The skill folder holds 1 other file (for example `durable-work-guidance.md`).

It sits in Agent Workflows, covering Subagents. The repository describes itself as: Make AI coding agents architecture-aware: baseline-first, evidence-verified, drift-checked, and safe across long tasks. The licence is MIT.

When your agent uses it

  • A task is multi-step
  • May span context resets
  • Risks losing state before completion

Example prompts

  • “/long-task-continuation”

Workflow steps

5 steps, taken from the first numbered list in SKILL.md.

  1. Capture requested outcome, scope, non-goals, risks, parent plan/goal, success
  2. Identify required baseline refs and record acknowledged, cited, and missing
  3. Choose inline or durable state once, then record the todo map, active slice,
  4. When an Execution Readiness View exists, retain its intent lock, scope
  5. For a new helper-backed record, use durable-work-guidance.md to create and

What it can do on your machine

Read from SKILL.md and the folder at commit 4edf34e. It shows what the files ask for, not the result of running them.

  • Tool permissions

    Pre-approves nothing: there is no allowed-tools line, so your agent's usual permission prompts apply.

    From allowed-tools in the SKILL.md frontmatter.

  • Runs code

    No scripts in the folder and no shell commands in SKILL.md.

    From the folder's file list and the shell code blocks in SKILL.md.

  • Network

    No URLs in SKILL.md.

    From URLs in SKILL.md, links to its own repository left out.

  • Credentials

    Names no API keys, tokens, secrets or passwords.

    From names ending in _API_KEY, _TOKEN, _SECRET, _KEY or _PASSWORD in SKILL.md.

Context cost

Long Task Continuation loads about 2.3k tokens when it runs. Until then it costs about 37 tokens; SKILL.md has 1,162 words of instructions outside code blocks.

Always · name and description, kept in context so the agent knows when to use it
~37
When it runs · the whole SKILL.md, loaded when a task matches
~2.3k

Estimates: characters ÷ 4, the usual rule of thumb; real counts depend on the model's tokenizer. Scripts and assets cost tokens only if the agent reads them.

Safety

Auto-check passed

The automated check found no risky patterns in SKILL.md.

Automated static check — not a guarantee. Review scripts before installing. It scans the text of SKILL.md for risky patterns (piping downloads into a shell, reading credential files, hidden Unicode, destructive commands); files beside SKILL.md are not scanned.

SKILL.md

The full file from GanyuanRan/Aegis at commit 4edf34e, republished under its MIT licence (© GanyuanRan). 1,162 words, ~2,343 tokens.

Download SKILL.mdSave it as .claude/skills/long-task-continuation/SKILL.md (or your agent's skills folder). This skill also uses 1 other file; get the full folder from GitHub.
name
long-task-continuation
description
Use when a task is multi-step, may span context resets or sessions, uses subagents, or risks losing state before completion.

Long Task Continuation

Overview

Keep long work checkpointed, resumable, drift-aware, and evidence-gated. This protocol does not execute plans, dispatch subagents, run tests, or grant completion authority.

Authority Boundary

The Method Pack owns continuation discipline only. It does not own the parent plan, host retry/watchdog behavior, authoritative GateDecision, evidence sufficiency, requirement acceptance, or completion.

When To Use

Use this skill when the work has meaningful phases, may be compacted/resumed or handed off, uses subagents, or explicitly needs continuity and drift control. Architecture, contract, shared-workflow, and verification-gate changes also benefit from it. Do not force it onto a short answer or one-command check.

Choose exactly one state carrier:

  • use a durable work/ record for medium+ work that actually crosses sessions, needs handoff, or requires resumable state;
  • otherwise keep one inline checkpoint.

Multi-step, todo-driven, possible-compaction, and subagent use do not force durable records by themselves. Do not create both carriers or a record per slice.

Required Artifacts

A durable task has one process trail under docs/aegis/work/YYYY-MM-DD-<slug>/. It keeps logical intent/baseline state, the latest todo/checkpoint/resume state, terminal evidence/drift state, and a completion reflection when warranted. These are TaskIntentDraft, BaselineReadSetHint, BaselineUsageDraft, ImpactStatementDraft, TodoCheckpointDraft, ResumeStateHint, DriftCheckDraft, and EvidenceBundleDraft views—not authoritative runtime records or separate plan owners.

Read only the lifecycle-matched section of durable-work-guidance.md:

  • ## Required Artifact Layout and ## Create A Durable Work Record for a new durable work record;
  • ## Update A Slice when an existing helper-backed record needs sidecar updates;
  • ## Retry Convergence Detail for retry/attempt bookkeeping;
  • ## Pause, Handoff, And Completion Bundle when preparing a pause, handoff, or completion bundle; and
  • ## Expanded State Fields only when natural checkpoint prose is ambiguous.

The reference owns artifact layout and <aegis-workspace-helper> command detail; this file owns carrier selection, resume order, drift decisions, and stop conditions.

An Execution Readiness View may be kept in the intent or active checkpoint for medium/high, handoff-prone, long-running, subagent-driven, architecture, contract, compatibility, or retirement-sensitive work. It renders existing intent, scope, baseline, owner, test, review, and drift constraints; it is not a new JSON artifact or completion authority.

Planless Slice Lane:

  • When an existing parent plan/spec owns a bounded task, reuse it and the current checkpoint. For a no-parent direct bounded request with no new durable or unclear verification boundary, use an inline checkpoint.
  • State one compact Slice Card: Goal, Parent plan/spec (or none — direct bounded request), Files, Boundary, Verification, and Stop.
  • The slice goal closes only that slice. Final completion returns to the parent or direct bounded request through verification-before-completion.
  • Do not create a plan/spec merely to give a micro-slice a parent, and do not create per-slice plans/specs or work records.
  • Escalate when a new owner, contract, schema, public API, architecture, migration, persistence, security/permission, distribution/release surface, unclear verification boundary, or mismatch with parent scope or acceptance appears.

When durable architecture decisions are in scope, these work records are the preferred ADR Auto Backfill source. Preserve decision signals, source refs, alternatives, compatibility, retirement, drift, and baseline-sync questions.

Start Protocol

Before execution:

  1. Capture requested outcome, scope, non-goals, risks, parent plan/goal, success evidence, and stop states (done | blocked | needs-verification | scope-exceeded).
  2. Identify required baseline refs and record acknowledged, cited, and missing refs. Missing authority pauses in needs-baseline-readback.
  3. Choose inline or durable state once, then record the todo map, active slice, completed slices/evidence, blockers, next step, and current branch/HEAD.
  4. When an Execution Readiness View exists, retain its intent lock, scope fence, baseline lock, compatibility/retirement boundary, tests, reviews, evidence, and rewind rules.
  5. For a new helper-backed record, use durable-work-guidance.md to create and structurally check it before implementation.

Retry Convergence Protocol

A failed verification is another attempt in the current slice, not a new slice. Keep failed-attempt telemetry out of terminal evidence and normal commits. Only evidence-finalized, blocked, or abandoned is terminal. When retry state reaches process-artifact-pressure, stop auto-retry and route to systematic-debugging or verification-before-completion. Load the durable reference only when the attempt/evidence commands or sidecar rules are needed.

Show full SKILL.md (522 more words)Show less

Per-Slice Protocol

Before each slice, state the current goal/todo, intended edits, explicit non-edits, verification, and readiness alignment. A bounded parent-plan or no-parent slice uses the compact Slice Card rather than a new plan/spec.

After each slice, update completed todos, evidence refs, newly used baseline refs, blockers, next step, and drift decision. When an active helper-backed work record exists, read durable-work-guidance.md and update that same record; never create another workstream for bookkeeping.

When patch-shape/ripple triage, an H-class finding, or a bounded compatibility mitigation fired, a locally green result does not clear the direction. Retain PatchShape, CanonicalOwner, UpwardDrillSignal, latest outcome, and one bounded evidence ref; do not copy raw logs or full diffs. If no fresh evidence exists, the state is needs-verification or partial.

Resume Protocol

Resume in this order:

  1. Read original intent, parent plan/goal, latest checkpoint and resume hint.
  2. Re-read required baseline refs and relevant active CONTEXT.md language.
  3. Read the Execution Readiness View when present.
  4. Compare checkpoint branch/HEAD, completed commits, evidence refs, and claims with the current worktree. Treat inherited completion claims as unverified until the worktree confirms them. Correct contradictions in the checkpoint, disclose them in the final report, and do not republish stale claims as current. Preserve the claim's subject, scope and resume-time state; distinguish that from later repair and from historical execution you cannot verify. One accurate correction may cover related claims. The record must adopt it, without elsewhere denying it or assigning it to another scope. A reviewer's classification is contestable: if material disagreement remains unresolved after checking evidence, retain needs-verification without forcing assent. If evidence resolves it, retain the supported conclusion and verify normally.
  5. Compare the active slice against intent lock, scope fence, baseline lock, compatibility/retirement boundary, tests, reviews, and non-goals.
  6. Re-run the drift decision, then name the next smallest authorized action.

Any disagreement among plan, checkpoint, baseline, context, readiness view, or worktree pauses execution. A semantic conflict routes to establishing-project-context; an unplanned repair re-reads the retained invariant, owner seam, patch shape, and causal topology, then route comparison to systematic-debugging. A new carrier name alone does not prove a new direction. Never resume from memory alone.

Drift Check

Check original intent and stop condition, parent scope/acceptance, compatibility, new owners/fallbacks/adapters/branches, retirement, evidence freshness, and any readiness locks. Allowed decisions are continue, pause-for-user, needs-baseline-readback, needs-verification, and blocked.

Never emit gate-passed, completion-granted, or authoritatively-safe.

Completion Candidate Protocol

Before a completion claim:

  1. Use aegis:verification-before-completion.
  2. Confirm every todo has status, blockers are resolved/externalized, evidence covers acceptance, and drift has no blocking state.
  3. If a durable record exists, load durable-work-guidance.md for the completion bundle and structural workspace check.
  4. For durable architecture work, pass the work record, proof bundle and ADR signals to verification for ADR Backfill Check.

Generated packs are future-runtime inputs only. Method Pack output remains verified evidence and advisory judgment, not authoritative completion.

Minimal Reporting Shape

Report naturally and omit empty structures. Keep these semantic slots visible: Aegis Visibility; current todo/active/completed/next; baseline usage decision; readiness state when present; fresh evidence; inherited-claim corrections when a resumed record overstated its state; retry/convergence state when relevant; drift decision; risk/unknown; and the next smallest safe action.

© GanyuanRan, MIT. Rendered from Markdown: HTML in the file is shown as text, images as links, and headings moved down two levels. Raw file

Files

SKILL.md and 1 other file in skills/long-task-continuation of GanyuanRan/Aegis.

  • SKILL.md
  • durable-work-guidance.md

Open the folder on GitHubat commit 4edf34e

Used in 1 other repository

We found 1 copy of this SKILL.md (exact, near-identical or edited) in other folders, from 1 other GitHub owner. This page covers the copy in GanyuanRan/Aegis, which our catalogue first saw on October 7, 2026.

Compare with similar skills

Long Task Continuation next to the 5 skills that share the most tags, products or categories with it. Stars are the repository's; “used in” counts other GitHub owners with a copy.

Long Task Continuation compared with similar skills
SkillStarsUsed inTokensAuto-checkLicenceRepo updated
Long Task Continuation this skillGanyuanRan/Aegis1.3k1 repos~2.3kAutomated safety check: PassMIT
Claude Code Agent Developmentanthropics/claude-plugins-official38k8 repos~2.8kAutomated safety check: PassApache-2.0
Subagent Driven DevelopmentAsvarox/allkaraoke26138 repos~1.2kAutomated safety check: PassNone
Dispatching Parallel Agentsultralisp/ultralisp25841 repos~1.5kAutomated safety check: PassNone
Paseo Advisor Second Opiniongetpaseo/paseo20k1 repos~756Automated safety check: PassCustom licence
Task Observerrebelytics/one-skill-to-rule-them-all3.2k1 repos~12kAutomated safety check: PassCC-BY-4.0

Similar skills

  • Claude Code Agent Development

    anthropics/claude-plugins-official

    Official

    Explains how to write agents for Claude Code plugins: the markdown file with YAML frontmatter, trigger descriptions, model and color settings, and system prompt design.

    38k GitHub starsUsed in 8 repos~2.8k tokens
    Agent WorkflowsAuto-check passed
  • Subagent Driven Development

    Asvarox/allkaraoke

    A skill your agent uses when executing implementation plans with independent tasks in the current session

    261 GitHub starsUsed in 38 repos~1.2k tokens
    Agent WorkflowsAuto-check passed
  • Dispatching Parallel Agents

    ultralisp/ultralisp

    A skill your agent uses when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies

    258 GitHub starsUsed in 41 repos~1.5k tokens
    Agent WorkflowsAuto-check passed
  • Launches one separate agent through Paseo to give a second opinion on the current task, with a self-contained briefing and no permission to edit files.

    20k GitHub starsUsed in 1 repo~756 tokens
    Agent WorkflowsAuto-check passed
  • Task Observer

    rebelytics/one-skill-to-rule-them-all

    Monitors task execution for skill improvement opportunities.

    3.2k GitHub starsUsed in 1 repo~12k tokens
    Agent WorkflowsAuto-check passed
  • O2 Review Loop

    openobserve/openobserve

    Splits a change into planner, coder and independent reviewer roles: you confirm a spec, a subagent implements it, and a separate reviewer checks each round's local WIP commit.

    22k GitHub stars~3.7k tokensUpdated today
    Agent WorkflowsAuto-check passed

More from GanyuanRan/Aegis

All 21 skills in this repo
  • Update Aegis

    GanyuanRan/Aegis

    A skill your agent uses when the user says aegis:update, asks to update or upgrade an installed Aegis method-pack, wants the latest Aegis version, or asks whether Aegis is current on this host.

    1.3k GitHub starsUsed in 2 repos~1.4k tokens
    Auto-check passed
  • Anti Entropy Governance

    GanyuanRan/Aegis

    A skill your agent uses when touching retiring old logic, collapsing duplicate owners, removing fallbacks, or schema/persistence/source-of-truth boundaries; identify opportunities automatically…

    1.3k GitHub starsUsed in 1 repo~3.1k tokens
    Auto-check passed
  • Executing Plans

    GanyuanRan/Aegis

    A skill your agent uses when executing a written implementation plan across sessions or with review checkpoints.

    1.3k GitHub starsUsed in 1 repo~2.3k tokens
    Auto-check passed
  • First Principles Review

    GanyuanRan/Aegis

    A skill your agent uses when asked for first-principles or Occam's-razor review, or when high-risk decisions involve competing constraints, fallback growth, duplicate owners, or architecture…

    1.3k GitHub starsUsed in 1 repo~2.5k tokens
    Auto-check passed
  • Goal Framing

    GanyuanRan/Aegis

    A skill your agent uses when the user explicitly sets an Aegis goal with /aegis-goal, Aegis goal:, or asks to define goal, success evidence, stop condition, or task boundaries before work.

    1.3k GitHub starsUsed in 1 repo~1.2k tokens
    Auto-check passed
  • A skill your agent uses when the user asks to establish shared project language, or project work exposes a conflicting, renamed, or deprecated domain term that needs active semantic modeling.

    1.3k GitHub starsUsed in 1 repo~1.5k tokens
    Auto-check: warnings

Categories

Questions about Long Task Continuation

What does Long Task Continuation do?

A skill your agent uses when a task is multi-step, may span context resets or sessions, uses subagents, or risks losing state before completion. Long Task Continuation is an agent skill from GanyuanRan/Aegis. Use when a task is multi-step, may span context resets or sessions, uses subagents, or risks losing state before completion.

When should I use Long Task Continuation?

Long Task Continuation fits situations like: A task is multi-step; may span context resets; risks losing state before completion.

How do I install Long Task Continuation in Claude Code?

Run `npx skills add GanyuanRan/Aegis --skill long-task-continuation -a claude-code`. Or copy the skill folder (skills/long-task-continuation in GanyuanRan/Aegis) into .claude/skills/long-task-continuation in your project. Claude Code loads it when a task matches its description.

How do I install Long Task Continuation in Codex?

Run `npx skills add GanyuanRan/Aegis --skill long-task-continuation -a codex`. Or copy the skill folder (skills/long-task-continuation in GanyuanRan/Aegis) into .agents/skills/long-task-continuation in your project. Codex loads it when a task matches its description.

Can I use Long Task Continuation in Cursor, Gemini CLI or GitHub Copilot?

Cursor, Gemini CLI, GitHub Copilot and OpenCode also load SKILL.md folders. With the skills CLI, run `npx skills add GanyuanRan/Aegis --skill long-task-continuation -a cursor` (or -a gemini-cli, github-copilot or opencode for the others). To copy it by hand, put the folder in .cursor/skills/long-task-continuation, .gemini/skills/long-task-continuation, .github/skills/long-task-continuation and .opencode/skills/long-task-continuation in your project.

What does Long Task Continuation need to run?

SKILL.md names no scripts, command-line tools or credentials: Long Task Continuation is instructions for the agent only.

Does Long Task Continuation access the network?

SKILL.md contains no URLs. Any network use would come from the scripts or tools the agent runs. This is read from the text; nothing was executed.

Is Long Task Continuation safe to install?

Our automated static check of SKILL.md found no risky patterns, such as piping downloads into a shell, reading credential files or hidden Unicode. It is not a guarantee. Review the folder before installing.

What licence does Long Task Continuation use?

Long Task Continuation is published under the MIT licence (the repository's licence). It allows redistribution, so the full SKILL.md is shown on this page.

How many tokens does Long Task Continuation use?

About 2.3k tokens (SKILL.md is roughly 9.4k characters). Agents keep only the skill's name and description in context until a task matches; then they load SKILL.md in full.

What are the alternatives to Long Task Continuation?

Skills that share tags, products or a category with Long Task Continuation: Claude Code Agent Development (anthropics/claude-plugins-official, 38k stars), Subagent Driven Development (Asvarox/allkaraoke, 261 stars), Dispatching Parallel Agents (ultralisp/ultralisp, 258 stars) and Paseo Advisor Second Opinion (getpaseo/paseo, 20k stars). The comparison table on this page puts their stars, adoption, token cost, safety result and licence side by side.

Who maintains Long Task Continuation?

GanyuanRan (a GitHub user) maintains it in GanyuanRan/Aegis, which has 1,327 GitHub stars. The repository holds 21 skills in this directory. The repository was last updated on October 3, 2026.

Source: GanyuanRan/Aegis on GitHub. Facts on this page come from the repository at the commit we read; the author's words are quoted as theirs.