[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"project-11487":3},{"id":4,"name":5,"fullName":6,"owner":7,"repo":5,"description":8,"homepage":9,"htmlUrl":10,"language":11,"languages":10,"totalLinesOfCode":10,"stars":12,"forks":13,"watchers":14,"openIssues":15,"contributorsCount":16,"subscribersCount":16,"size":16,"stars1d":17,"stars7d":18,"stars30d":19,"stars90d":16,"forks30d":16,"starsTrendScore":20,"compositeScore":21,"rankGlobal":10,"rankLanguage":10,"license":22,"archived":23,"fork":23,"defaultBranch":24,"hasWiki":25,"hasPages":23,"topics":26,"createdAt":10,"pushedAt":10,"updatedAt":42,"readmeContent":43,"aiSummary":44,"trendingCount":16,"starSnapshotCount":16,"syncStatus":45,"lastSyncTime":46,"discoverSource":47},11487,"Aegis","GanyuanRan\u002FAegis","GanyuanRan","Make AI coding agents architecture-aware: baseline-first, evidence-verified, drift-checked, and safe across long tasks.","",null,"Shell",490,26,129,1,0,36,88,318,108,94.29,"MIT License",false,"main",true,[27,28,29,30,31,32,33,34,35,36,37,38,39,40,41],"add","agent-skills","ai-agents","ai-coding","architecture-driven-development","baseline-first","claude-code","codex","coding-agents","developer-tools","evidence-driven","opencode","software-architecture","tdd","tlref","2026-06-12 04:00:55","\u003Cp align=\"center\">\n    \u003Ca href=\"https:\u002F\u002Flinux.do\u002Ft\u002Ftopic\u002F2108966\u002F20\" alt=\"LINUX DO\">\n        \u003Cimg\n            src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FLINUX-DO-FFB003.svg?logo=data:image\u002Fsvg%2bxml;base64,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\" \u002F>\u003C\u002Fa>\n\u003C\u002Fp>\n\n\u003Cp align=\"center\">\n    \u003Cimg src=\"assets\u002Faegis-hero.png\" alt=\"Aegis architecture-driven AI coding agent hero banner\" \u002F>\n\u003C\u002Fp>\n\n\u003Cp align=\"center\">\n    \u003Ca href=\"https:\u002F\u002Fgithub.com\u002FGanyuanRan\u002FAegis\">⭐ Star the repo here to help accelerate updates ❤️\u003C\u002Fa>\n\u003C\u002Fp>\n\n# Aegis\n\n\u003Cp align=\"center\">\n    \u003Cstrong>Aegis Method Pack\u003C\u002Fstrong>\u003Cbr\u002F>\n    Runtime-ready workflow discipline for AI coding agents.\n\u003C\u002Fp>\n\n\u003Cp align=\"center\">\n    \u003Ca href=\"README.md\">\u003Cstrong>English\u003C\u002Fstrong>\u003C\u002Fa>\n    ·\n    \u003Ca href=\"README.zh-CN.md\">\u003Cstrong>中文\u003C\u002Fstrong>\u003C\u002Fa>\n    ·\n    \u003Ca href=\"docs\u002Fcurrent\u002FAEGIS_WORKFLOW_GUIDE.md\">Workflow Guide\u003C\u002Fa>\n    ·\n    \u003Ca href=\"docs\u002Fcurrent\u002FAEGIS_WORKFLOW_GUIDE_ZH.md\">工作流程说明\u003C\u002Fa>\n\u003C\u002Fp>\n\n## Minimal Install\n\nIf you are using an AI coding agent, you can ask it to install Aegis for you:\n\n```text\nPlease read the installation instructions in https:\u002F\u002Fgithub.com\u002FGanyuanRan\u002FAegis carefully, identify the correct path for my AI coding host, install Aegis globally, restart or reload the host if needed, then run complete-install verification from the Aegis method-pack root with `python scripts\u002Faegis-doctor.py --write-config --json`. Treat the install as complete only if the JSON output includes `\"ok\": true`, `\"workspaceSupport\": \"available\"`, and `\"configStatus\": \"configured\"`; if the host uses a separate skill discovery directory, also verify it with `--discovery-root \u003Cpath>`.\n```\n\n## Updating Aegis\n\nIf Aegis is already installed, you can ask your AI coding agent to update it for you:\n\n```text\nPlease update my installed Aegis to the latest main branch version from https:\u002F\u002Fgithub.com\u002FGanyuanRan\u002FAegis, using the correct update path for my current AI coding host, then restart or reload the host if needed and run complete-install verification from the Aegis method-pack root with `python scripts\u002Faegis-doctor.py --write-config --json`. Treat the update as complete only if the JSON output includes `\"ok\": true`, `\"workspaceSupport\": \"available\"`, and `\"configStatus\": \"configured\"`; if the host uses a separate skill discovery directory, also verify it with `--discovery-root \u003Cpath>`.\n```\n\n## Optional Lite Global Rules\n\nFor smoother host-level behavior, copy the whole block below into your AI\ncoding tool's global user rules. It improves Aegis routing and skill triggering\nwithout duplicating the full workflows:\n\n```markdown\n# Aegis Lite Global Rules\n\nIf Aegis is installed:\n\n- At the start of each turn, check whether the task matches an installed Aegis\n  skill. If it matches, load and follow that skill.\n- Simple, local, low-risk tasks may use a fast path. Do not expand the full\n  governance workflow just because Aegis exists.\n- Complex, diagnostic, architecture, refactor, contract, cross-module, shared\n  module, compatibility, or long-running tasks should use the relevant Aegis\n  workflow by default.\n- Before implementation, identify the goal, scope, impact surface, and\n  verification method. Read project baseline or authority docs when relevant.\n- Before claiming completion, provide fresh verification evidence. If\n  verification is blocked, state the blocker and residual risk.\n- Aegis is a method layer, not a final authority system. Do not claim final\n  gate decisions or completion authority.\n- The user's current instruction and the target project's rules take priority\n  over Aegis guidance.\n```\n\nFor stricter teams or governance-heavy projects, you can instead start from the\nfull advanced templates and merge only the parts you need:\n\n- [Advanced English template](GLOBAL_USER_RULES_TEMPLATE.md)\n- [Advanced Chinese template](GLOBAL_USER_RULES_TEMPLATE.zh-CN.md)\n\n## Activation Mode\n\nAegis defaults to automatic mode. To switch to manual mode, edit:\n\n```text\n~\u002F.config\u002Faegis\u002Fconfig.toml\n```\n\nWindows:\n\n```text\n%USERPROFILE%\\.config\\aegis\\config.toml\n```\n\nIf the file does not exist, create it manually. Add:\n\n```toml\nactivation_mode = \"explicit\"\n```\n\nTo return to automatic mode, set `activation_mode = \"auto\"` or delete the file.\n\nThen restart the host. In explicit mode, supported bootstrap hooks stop\ninjecting Aegis automatically, while direct skill calls such as\n`aegis:using-aegis` remain available. Host caveats are documented in\n[docs\u002Fcurrent\u002FAEGIS_ACTIVATION_MODE.md](docs\u002Fcurrent\u002FAEGIS_ACTIVATION_MODE.md).\n\n`Aegis` is an Architecture-Driven Development (ADD) method pack for AI coding agents.\n\nIt builds on the original `superpowers` methodology and adds evidence-driven governance, TLREF execution flow, and dual-track repair\u002Fretirement rules.\n\nIn Aegis, ADD means the agent should understand the project's baseline, architecture boundaries, owners, impact surface, compatibility constraints, and verification path before making substantial changes.\n\nCurrent release shape:\n\n> `Aegis Method Pack (runtime-ready)`\n\nThis repository is **not** the full `Aegis Platform`. It does not provide authoritative runtime-core decisions, authoritative `GateDecision`, or completion authority.\n\n## Problems Aegis Solves\n\nAI coding agents are strong at local execution, but long or risky software work often fails for process reasons:\n\n- work starts before the task boundary and baseline are clear\n- the agent claims completion without fresh evidence\n- bug fixes add new fallback logic while old owners keep running\n- long tasks lose state after compaction, handoff, or multi-agent work\n- architecture drift is noticed only after the change has already spread\n\nAegis addresses those problems at the method-pack layer. It makes the agent frame the task,\nread the relevant baseline, keep evidence close to claims, track repair and retirement together,\nand maintain resumable checkpoints for long work.\n\n## What Users Get\n\nInstalling Aegis gives an AI coding tool a stricter development discipline without requiring a new runtime platform:\n\n- clearer task framing before edits\n- safer debugging and refactoring loops\n- fewer unsupported \"done\" claims\n- better long-task continuity through todo checkpoints, resume hints, and drift checks\n- explicit compatibility and retirement thinking when behavior changes\n- portable workflows across Codex, OpenCode, Claude Code, and other skill-aware hosts\n\n## What Aegis Adds\n\nAegis keeps the useful parts of `superpowers`:\n\n- composable skills\n- skill-triggered workflows\n- multi-host install and plugin distribution skeletons\n- implementation planning, review, debugging, and verification practices\n\nAegis adds a stricter governance spine:\n\n- baseline-first work\n- evidence before claims\n- impact-aware task framing\n- TLREF \u002F DIVE \u002F Reflection \u002F QA execution discipline\n- repair track plus retirement track for bug fixes, refactors, contract changes, and governance cleanup\n- long-task continuation with todo checkpoints, resume hints, drift checks, and evidence bundles\n- runtime-ready artifacts that remain drafts, hints, projections, and evidence bundles\n\n## Current Scope\n\nAegis currently owns:\n\n- method-pack skills\n- initial instructions and contributor guardrails\n- host installation guidance\n- representative tests and verification assets\n- runtime-ready artifact shapes\n- release, rollback, known-limitation, and compatibility checklists for maintainers\n\nAegis does not currently own:\n\n- `Host Adapters`\n- `Runtime Core`\n- authoritative `GateDecision`\n- final completion authority\n- full production rollout guarantees\n\nFor the current authority map, read:\n\n- [docs\u002Fcurrent\u002FREADME.md](docs\u002Fcurrent\u002FREADME.md)\n- [docs\u002Fadr\u002FADR-0001-aegis-method-pack-is-not-runtime-core.md](docs\u002Fadr\u002FADR-0001-aegis-method-pack-is-not-runtime-core.md)\n\n## Host Compatibility\n\nAegis keeps the multi-host plugin-installable goal.\n\nCurrent host-facing status:\n\n| Host | Current status |\n| --- | --- |\n| `Codex` | Representative smoke path verified; Git Bash naive smoke still has known observation items |\n| `OpenCode` | Base suite and integration closeout have passed in the current method-pack scope |\n| `Claude Code` | Plugin skeleton and install guide exist; release-level fresh host smoke is still pending |\n| `CodeBuddy` | Plugin skeleton and native `SKILL.md` manual install guide exist; release-level fresh host smoke is still pending |\n| `DeepSeek-TUI` | Native `SKILL.md` discovery supports manual Aegis skill install; release-level fresh host smoke is still pending |\n| `Trae` | Native `SKILL.md` discovery supports manual Aegis skill install; release-level fresh host smoke is still pending |\n\nOther hosts remain product targets, but are not yet current release-level verdicts.\n\nRead:\n\n- [docs\u002Fcurrent\u002FAEGIS_HOST_COMPATIBILITY_MATRIX_SNAPSHOT.md](docs\u002Fcurrent\u002FAEGIS_HOST_COMPATIBILITY_MATRIX_SNAPSHOT.md)\n- [docs\u002Fcurrent\u002FAEGIS_KNOWN_LIMITATIONS.md](docs\u002Fcurrent\u002FAEGIS_KNOWN_LIMITATIONS.md)\n\n## Installation\n\nAegis can be installed through each host's native discovery or plugin path.\nA public marketplace listing is not required for the paths below.\n\nAfter installation and host restart, Aegis skills are discovered automatically.\nFor normal use, users can ask for development work naturally; the agent should\nselect the relevant Aegis method when the task matches a skill. Explicit skill\ncommands are still available when you want to force, test, or debug a workflow.\n\n### Codex\n\nmacOS \u002F Linux:\n\n```bash\ngit clone https:\u002F\u002Fgithub.com\u002FGanyuanRan\u002FAegis.git ~\u002F.codex\u002Faegis\nmkdir -p ~\u002F.agents\u002Fskills\nln -s ~\u002F.codex\u002Faegis\u002Fskills ~\u002F.agents\u002Fskills\u002Faegis\n```\n\nWindows PowerShell:\n\n```powershell\ngit clone https:\u002F\u002Fgithub.com\u002FGanyuanRan\u002FAegis.git \"$env:USERPROFILE\\.codex\\aegis\"\nNew-Item -ItemType Directory -Force -Path \"$env:USERPROFILE\\.agents\\skills\"\ncmd \u002Fc mklink \u002FJ \"$env:USERPROFILE\\.agents\\skills\\aegis\" \"$env:USERPROFILE\\.codex\\aegis\\skills\"\n```\n\nOptional Codex config for subagent-heavy skills:\n\n```toml\n[features]\nmulti_agent = true\n```\n\nRestart Codex, then ask it to use `aegis:using-aegis` or an Aegis skill such as\n`brainstorming`.\n\n### OpenCode\n\nUse the plugin shortcut by adding Aegis to the `plugin` array in your global or\nproject `opencode.json`:\n\n```json\n{\n  \"plugin\": [\"aegis@git+https:\u002F\u002Fgithub.com\u002FGanyuanRan\u002FAegis.git\"]\n}\n```\n\nIf the file already has plugins, append Aegis:\n\n```json\n{\n  \"plugin\": [\n    \"other-plugin\",\n    \"aegis@git+https:\u002F\u002Fgithub.com\u002FGanyuanRan\u002FAegis.git\"\n  ]\n}\n```\n\nThen restart OpenCode and verify:\n\n```bash\nopencode --version\n```\n\nAsk: `Tell me about your aegis`.\n\n### Claude Code\n\nMarketplace flow:\n\n```bash\nclaude plugin marketplace add GanyuanRan\u002FAegis\nclaude plugin install aegis@aegis-dev --scope user\n```\n\nLocal checkout flow:\n\n```bash\ngit clone https:\u002F\u002Fgithub.com\u002FGanyuanRan\u002FAegis.git ~\u002Faegis\nclaude --plugin-dir ~\u002Faegis\n```\n\nInside Claude Code, run `\u002Freload-plugins`, then try `\u002Faegis:using-aegis`.\n\n### CodeBuddy\n\nCodeBuddy supports both plugin metadata and native `SKILL.md` skill discovery.\nFor the transparent manual path:\n\nmacOS \u002F Linux:\n\n```bash\ngit clone https:\u002F\u002Fgithub.com\u002FGanyuanRan\u002FAegis.git ~\u002F.codebuddy\u002Faegis\nmkdir -p ~\u002F.codebuddy\u002Fskills\ncp -R ~\u002F.codebuddy\u002Faegis\u002Fskills\u002F* ~\u002F.codebuddy\u002Fskills\u002F\n```\n\nWindows PowerShell:\n\n```powershell\ngit clone https:\u002F\u002Fgithub.com\u002FGanyuanRan\u002FAegis.git \"$env:USERPROFILE\\.codebuddy\\aegis\"\nNew-Item -ItemType Directory -Force -Path \"$env:USERPROFILE\\.codebuddy\\skills\"\nCopy-Item -Recurse -Force \"$env:USERPROFILE\\.codebuddy\\aegis\\skills\\*\" \"$env:USERPROFILE\\.codebuddy\\skills\\\"\n```\n\nAegis also ships `.codebuddy-plugin\u002F` metadata for CodeBuddy plugin flows.\nRestart CodeBuddy, then ask it to describe its Aegis skills.\n\n### DeepSeek-TUI\n\nDeepSeek-TUI discovers skills from `SKILL.md` directories. Install Aegis by\ncopying its skill directories into DeepSeek-TUI's global skills path:\n\nmacOS \u002F Linux:\n\n```bash\ngit clone https:\u002F\u002Fgithub.com\u002FGanyuanRan\u002FAegis.git ~\u002F.deepseek\u002Faegis\nmkdir -p ~\u002F.deepseek\u002Fskills\ncp -R ~\u002F.deepseek\u002Faegis\u002Fskills\u002F* ~\u002F.deepseek\u002Fskills\u002F\n```\n\nWindows PowerShell:\n\n```powershell\ngit clone https:\u002F\u002Fgithub.com\u002FGanyuanRan\u002FAegis.git \"$env:USERPROFILE\\.deepseek\\aegis\"\nNew-Item -ItemType Directory -Force -Path \"$env:USERPROFILE\\.deepseek\\skills\"\nCopy-Item -Recurse -Force \"$env:USERPROFILE\\.deepseek\\aegis\\skills\\*\" \"$env:USERPROFILE\\.deepseek\\skills\\\"\n```\n\nRestart DeepSeek-TUI, then verify with `\u002Fskills` and `\u002Fskill using-aegis`.\n\n### Trae\n\nTrae discovers skills from `SKILL.md` directories. Install Aegis by copying its\nskill directories into Trae's global skills path:\n\nmacOS \u002F Linux:\n\n```bash\ngit clone https:\u002F\u002Fgithub.com\u002FGanyuanRan\u002FAegis.git ~\u002F.trae\u002Faegis\nmkdir -p ~\u002F.trae\u002Fskills\ncp -R ~\u002F.trae\u002Faegis\u002Fskills\u002F* ~\u002F.trae\u002Fskills\u002F\n```\n\nWindows PowerShell:\n\n```powershell\ngit clone https:\u002F\u002Fgithub.com\u002FGanyuanRan\u002FAegis.git \"$env:USERPROFILE\\.trae\\aegis\"\nNew-Item -ItemType Directory -Force -Path \"$env:USERPROFILE\\.trae\\skills\"\nCopy-Item -Recurse -Force \"$env:USERPROFILE\\.trae\\aegis\\skills\\*\" \"$env:USERPROFILE\\.trae\\skills\\\"\n```\n\nRestart Trae, then ask it to describe its Aegis skills.\n\nFull host guides:\n\n- [Claude Code](docs\u002FREADME.claude-code.md)\n- [CodeBuddy](docs\u002FREADME.codebuddy.md)\n- [Codex](docs\u002FREADME.codex.md)\n- [DeepSeek-TUI](docs\u002FREADME.deepseek-tui.md)\n- [OpenCode](docs\u002FREADME.opencode.md)\n- [Trae](docs\u002FREADME.trae.md)\n\nThe project still preserves the broader multi-host distribution skeleton inherited from `superpowers`, including Cursor and Gemini-related package surfaces. Those surfaces should not be interpreted as current fresh release-level closeout unless the compatibility matrix says so.\n\n## First Project Baseline\n\nAegis works best when the target project has a small, explicit baseline before\nlarge development work starts. Without one, Aegis can still run, but task\nframing, authority lookup, verification, and drift checks rely more on ad-hoc\ncontext and may be less stable.\n\nFor a new or undocumented project, first ask the agent to create a lightweight\nproject baseline, for example:\n\n```text\nUse Aegis to establish this project's baseline before implementation:\npurpose, current architecture, authority docs, run\u002Ftest commands, compatibility\nboundaries, non-goals, and verification expectations.\n```\n\nFor an existing project, point the agent at the repository's current source of\ntruth, such as `README`, `CONTRIBUTING`, architecture docs, ADRs, or local\nproject rules, and ask it to treat those as baseline references before changing\ncode.\n\n## Execution Model\n\nAegis is not a daemon, background runner, or authoritative runtime core.\nIt works through host-level skill discovery, bootstrap context, and explicit skill loading.\n\nAutomatic behavior:\n\n- Codex discovers Aegis skills from the configured skills directory at startup.\n- OpenCode loads the Aegis plugin, mirrors skills into OpenCode's global skills path,\n  and injects compact bootstrap context.\n- Claude Code loads Aegis through its plugin namespace or a local plugin directory.\n- CodeBuddy discovers copied Aegis skill directories from native `SKILL.md`\n  skill paths, or loads Aegis through `.codebuddy-plugin\u002F` metadata.\n- DeepSeek-TUI discovers copied Aegis skill directories from its native\n  `SKILL.md` skill paths.\n- Trae discovers copied Aegis skill directories from its native `SKILL.md`\n  skill paths.\n- `using-aegis` tells the agent to check whether a task-specific skill applies before responding.\n- In day-to-day use, you do not need to manually name a skill for every request;\n  explicit commands are the override path when you want a specific method.\n\nExplicit use:\n\n- Ask for a skill by name, such as `aegis:brainstorming`,\n  `aegis:systematic-debugging`, `aegis:long-task-continuation`,\n  or `aegis:verification-before-completion`.\n- Use `\u002Faegis-goal \u003Ctask>` or portable `Aegis goal: \u003Ctask>` when you want a\n  thin goal frame before work starts. It sets goal, success evidence, stop\n  condition, and non-goals, then routes onward without creating project files\n  by default.\n  Example: `Aegis goal: Fix the auth refresh bug without rewriting the auth system.`\n- In OpenCode, use the native `skill` tool, for example: `use skill tool to load aegis\u002Fbrainstorming`.\n- In Claude Code, use the plugin namespace, for example: `\u002Faegis:using-aegis`.\n- In CodeBuddy, ask it to load an Aegis skill such as `systematic-debugging`.\n- In DeepSeek-TUI, use the native skill command, for example: `\u002Fskill systematic-debugging`.\n- In Trae, ask it to load an Aegis skill such as `systematic-debugging`.\n\nLong-task behavior:\n\n- Aegis can keep `TodoCheckpointDraft`, `ResumeStateHint`, `DriftCheckDraft`,\n  and `EvidenceBundleDraft` discipline around long work.\n- This improves resumability and reduces drift, but it does not create a host watchdog,\n  automatic retry loop, or final completion authority.\n\n## Core Workflow\n\nThe method pack is organized around agent workflows:\n\n1. **Brainstorming**\n   - clarify intent, scope, impact, and baseline read set before implementation\n2. **Writing Plans**\n   - produce bite-sized, verifiable plans with exact files, verification steps, and compatibility boundaries\n3. **Systematic Debugging**\n   - move from symptoms to root cause with evidence before fixes\n4. **Test-Driven Development**\n   - use red\u002Fgreen\u002Frefactor where applicable\n5. **Requesting Code Review**\n   - review for behavioral risks, regressions, and missing tests\n6. **Verification Before Completion**\n   - no completion claim without fresh verification evidence\n\nAegis routes work by complexity before implementation:\n\n- Low-complexity tasks can proceed with a concise intent, baseline check, TDD, and verification.\n- Medium-complexity tasks require a baseline read set, a Spec Brief or stable requirements, a plan, and atomic tasks before TDD.\n- High-complexity tasks require a Design Spec and plan first, with user review where the workflow calls for it.\n\nWorkflow Quality guardrails keep that routing practical: simple tasks stay on\nthe fast path, medium\u002Fhigh-risk tasks get the right evidence and artifacts, and\noutputs use compact contracts before expanding into full workflow structure.\nRead [docs\u002Fcurrent\u002FAEGIS_WORKFLOW_QUALITY_BASELINE.md](docs\u002Fcurrent\u002FAEGIS_WORKFLOW_QUALITY_BASELINE.md).\n\nWhen a project needs persistent Aegis records, Aegis creates a lightweight\nproject workspace lazily. The default workspace includes `README.md`,\n`INDEX.md`, `BASELINE-GOVERNANCE.md`, and standard `adr\u002F`, `baseline\u002F`,\n`specs\u002F`, `plans\u002F`, and `work\u002F` directories under `docs\u002Faegis\u002F`. Task process\nrecords live under `docs\u002Faegis\u002Fwork\u002FYYYY-MM-DD-\u003Ctask-slug>\u002F`. Existing project\ndocs and ADRs remain the preferred authority; reusable Aegis outputs are\npromoted only when the workflow needs them.\n\nRecommended installs keep the Aegis method-pack root available so project\nworkspace support can be verified after installation. The Aegis method-pack\nrepository itself does not ship a precreated live `docs\u002Faegis\u002F` workspace.\nWorkspace structure checks do not judge evidence sufficiency or grant\ncompletion authority.\n\nMaintainers can verify a complete install and write the durable local helper\npaths with:\n\n```bash\npython scripts\u002Faegis-doctor.py --write-config --json\n```\n\nComplete-install verification must report `\"ok\": true`,\n`\"workspaceSupport\": \"available\"`, and `\"configStatus\": \"configured\"`.\n\nWhen a host exposes a separate skill discovery directory, pass it as\n`--discovery-root \u003Cpath>` to confirm it resolves to the current method-pack\nskills rather than a stale copied version.\n\nIf Aegis is installed but the expected skill does not trigger automatically,\ntreat it as trigger-chain diagnosis rather than a prompt wording problem:\n\n1. verify the method-pack version and install root\n2. verify the host discovery directory points at the current `skills\u002F`\n3. confirm the host was restarted or reloaded when required\n4. check `activation_mode` and whether automatic bootstrap is expected\n5. explicitly invoke `aegis:using-aegis` and then the expected skill\n6. compare the task against the trigger-health matrix\n\nAfter a long session, heavy tool output, resume, or context compaction,\nexplicitly re-run `aegis:using-aegis` to refresh routing before continuing\nnon-trivial work.\n\nSee `docs\u002Fcurrent\u002FAEGIS_TRIGGER_HEALTH_BASELINE.md` for the diagnostic layers.\n\nFor bug fixes, architecture changes, contract work, and governance cleanup, Aegis requires:\n\n- **Repair track**\n  - real root cause\n  - canonical owner\n  - smallest necessary change\n  - compatibility boundary\n  - verification method\n- **Retirement track**\n  - old owner \u002F fallback \u002F patch location\n  - whether it is still active\n  - reason to keep it, if any\n  - deletion or convergence trigger\n  - validation before removal\n\n## Runtime-Ready Artifacts\n\nCurrent method-pack outputs may include:\n\n- `TaskIntentDraft`\n- `BaselineReadSetHint`\n- `ImpactStatementDraft`\n- `EvidenceBundleDraft`\n- `GateInputPack`\n- `SubagentContextPacket`\n- `TodoCheckpointDraft`\n- `ResumeStateHint`\n- `DriftCheckDraft`\n\nThese are advisory and runtime-ready. They are not authoritative runtime decisions.\n\nRead:\n\n- [docs\u002Fcurrent\u002FAEGIS_ARTIFACT_SCHEMA_BASELINE.md](docs\u002Fcurrent\u002FAEGIS_ARTIFACT_SCHEMA_BASELINE.md)\n- [docs\u002Fcurrent\u002FAEGIS_RUNTIME_READY_BOUNDARY.md](docs\u002Fcurrent\u002FAEGIS_RUNTIME_READY_BOUNDARY.md)\n\n## Testing\n\nPrimary verification entry:\n\n```bash\nbash tests\u002Fe2e\u002Frun-all.sh --full --host-profile fast\n```\n\nFocused checks:\n\n```bash\nbash tests\u002Fe2e\u002Fboundary-compliance-check.sh\nbash tests\u002Fe2e\u002Fartifact-schema-check.sh\nbash tests\u002Fe2e\u002Fworkflow-quality-check.sh\nbash tests\u002Fopencode\u002Frun-tests.sh\nbash tests\u002Fcodex-plugin-sync\u002Ftest-sync-to-codex-plugin.sh\n```\n\nRead:\n\n- [docs\u002Ftesting.md](docs\u002Ftesting.md)\n- [docs\u002Fcurrent\u002FAEGIS_METHOD_PACK_RELEASE_CHECKLIST.md](docs\u002Fcurrent\u002FAEGIS_METHOD_PACK_RELEASE_CHECKLIST.md)\n\n## Contributing\n\nRead:\n\n- [CONTRIBUTING.md](CONTRIBUTING.md)\n- [SECURITY.md](SECURITY.md)\n- [SUPPORT.md](SUPPORT.md)\n- [CODE_OF_CONDUCT.md](CODE_OF_CONDUCT.md)\n\nBefore changing behavior-shaping skill content, also read:\n\n- [skills\u002Ffirst-principles-review\u002FSKILL.md](skills\u002Ffirst-principles-review\u002FSKILL.md)\n- [skills\u002Fwriting-skills\u002FSKILL.md](skills\u002Fwriting-skills\u002FSKILL.md)\n- [skills\u002Fverification-before-completion\u002FSKILL.md](skills\u002Fverification-before-completion\u002FSKILL.md)\n\n## Relationship To Superpowers\n\nAegis is derived from **[Superpowers](https:\u002F\u002Fgithub.com\u002Fobra\u002Fsuperpowers)**, created by [Jesse Vincent](https:\u002F\u002Fgithub.com\u002Fobra). Superpowers pioneered the idea of composable, multi-harness agent skills — the foundation this project builds on.\n\nWe are grateful to Jesse and all Superpowers contributors for creating and maintaining the original project under the MIT license, and for establishing the plugin distribution patterns (Claude Code, Codex, Cursor, OpenCode, Gemini CLI) that Aegis continues to use.\n\nThis project adds a governance-focused method layer and public release path for the `Aegis Method Pack`, while preserving Superpowers' zero-dependency philosophy and multi-harness compatibility.\n\n## Acknowledgments\n\nWe thank [Matt Pocock](https:\u002F\u002Fgithub.com\u002Fmattpocock) and all contributors to [mattpocock\u002Fskills](https:\u002F\u002Fgithub.com\u002Fmattpocock\u002Fskills) (MIT license) for sharing their skill designs openly. Several ideas from that project — particularly around concise communication, shared language glossaries, and disciplined debugging — have influenced Aegis skill design.\n\n| Aegis skill | Inspired by | What we adapted |\n|-------------|-------------|-----------------|\n| `communicating-concisely` | `\u002Fcaveman` | Ultra-compressed communication mode with auto-clarity exception |\n| `establishing-project-context` | `\u002Fgrill-with-docs` | CONTEXT.md shared language system, terminology tightening during brainstorming |\n| ADR creation gate | `\u002Fgrill-with-docs` ADR discipline | Three-condition gate before creating architecture decision records |\n| Feedback loop construction | `\u002Fdiagnose` Phase 1 | Priority ladder for building automated bug reproduction loops |\n\nThese ideas were re-implemented in Aegis format — shorter, multi-harness compatible, and integrated with the TLREF\u002FDIVE\u002FReflection governance spine rather than copied verbatim.\n\nInternal implementation notes are kept out of the public release tree. The\npublic contract is the skill content, current authority docs, and this\nacknowledgment.\n\n## License\n\nMIT License. See [LICENSE](LICENSE).\n","Aegis 是一个旨在使 AI 编码代理具备架构感知能力的项目，通过基线优先、证据验证、漂移检查等机制确保长任务执行的安全性。其核心功能包括基于架构驱动的开发流程、持续集成与测试支持，以及对代码质量的严格把控。技术上，Aegis 采用 Shell 脚本实现，并遵循 MIT 许可协议开放源代码。适用于需要在复杂软件架构中利用 AI 自动化编码的场景，如大型企业级应用开发或开源项目维护，能够有效提升开发效率和代码安全性。",2,"2026-06-11 03:31:59","CREATED_QUERY"]