agent-memory
Persist durable agent metadata after self-review, user review, PR comments, or a correction.
Skill by sardul3
Install
Works with Claude Code, Claude apps, Codex, Cursor, Gemini CLI, GitHub Copilot, VS Code, OpenCode, Goose, Windsurf. Read the repository before you give any tool access to your data.
npx agentmarket install skill/sardul3-agent-memory
Asks which of your tools to add it to (Claude Code, Codex, Cursor, VS Code, Windsurf, Gemini CLI, Claude Desktop) and shows the change first. See https://agentmart.dev/cli.
Get the weekly diff
The weekly diff: one email a week with new agent tools, notable updates and grade changes, plus security alerts for tools you follow. Double opt-in, one-click unsubscribe.
About
Use cases: legal, agent-building
Reviews
Write a reviewNo reviews yet. Installed it? Reviews come from verified installers only. How reviews work
Permissions
What this skill can do and what it asks for, from its SKILL.md. Checked Oct 25, 2026.
- Tools
- No tools found
- Secrets
- None asked for
- Install scripts
- None found
Network hosts it talks to 1
codeload.github.com· found in install command
Files it touches
~/.claude/skills/agent-memory
What's inside
- Skill name
agent-memory- Description
- Persist durable agent metadata after self-review, user review, PR comments, or a correction. Use when verdict.json metadata[], review feedback, or a user correction implies a missing convention, term, or stack law.
- License
- MIT
- Instructions
- 63 lines in SKILL.md · view on GitHub
Before installing
- Instructions read files in your home folder (~/.ssh, ~/.aws…)
Files (1)
- SKILL.md
SKILL.md (first lines)
# Agent memory
Feedback is wasted if the next session repeats the same miss. Prefer the **harness** over hand-editing.
## Structured path (preferred)
1. On self-review, put durable misses in `verdict.json` as `metadata[]` (reviewer may only flag; parent may fill):
```json
{
"verdict": "good",
"summary": "one sentence",
"risks": [],
"metadata": [
{
"target": "agents",
"text": "Isolation default is native git worktree; treehouse is opt-in.",
"reason": "review assumed treehouse"
},
{
"target": "rule",
"path": ".claude/rules/python-tests.md",
"globs": ["**/*.py"],
"text": "Prefer pytest for new packages.",
"reason": "PR comment"
}
]
}
```
2. Conductor auto-applies when `agent_memory.auto_apply: true` (default), writing `memory-applied.json` in the run dir.
3. Manual: `python3 ~/.claude/hooks/dev-loop/cli.py agent-memory --repo REPO --verdict PATH` (or `--key TICKET`).
Targets: `agents` | `rule` | `context` | `readme`. Rules must be path-scoped (`globs`, never `always_apply`). Secrets/hooks paths are rejected. `durable: false` is rejected.
## When (still apply by hand if no verdict)
- User/PR review corrects how agents should work and you are not going through `verdict.json`.
- A domain term or hard-to-reverse decision was resolved (`domain-glossary`).
Skip: one-ticket AC, style nits, speculative rewrites, secrets, SessionStart text.
## Where (pick one)
| Signal | File | Rule |
| ------ | ---- | ---- |
| Convention every future agent on **this repo** must keep | `AGENTS.md` / `CLAUDE.md` | Harness appends under `## Agent memory`. |
| Stack law missing for a path | path-scoped `.claude/rules` / `.cursor/rules` | `alwaysApply: false` + `globs`. |
| Domain term | `CONTEXT.md` via `domain-glossary` | Glossary only. |
| Hard to reverse + surprising + real trade-off | `docs/adr/` via `domain-glossary` | Skip unless all three. |
| Install / usage / public API would be wrong | `README.md` | That case only. |
Reviewer is read-only on metadata files. Parent writes `metadata[]` or calls the CLI.
## Do not
- Invent ADRs or always-on rules.
- Stow notes into SessionStart or CLAUDE.md tax.
- Update conductor secrets, hooks, or `~/.config/dev-conductor/secrets.env`.
- Copy skill bodies into AGENTS.md.
- Treat `SPEC_APPROVED` as ticket-done or chat as the source of Jira AC.From sardul3/dev-conductor, MIT.
More from sardul3/dev-conductor
- 0.524score
session-handoff
sardul3
Compact this conversation into a handoff file for a fresh agent.
SkillBFreshness: BClaude Code +9 - 0.524score
design-tree-interview
sardul3
Relentless design-tree interview.
SkillBFreshness: BClaude Code +9 - 0.524score
dev-loop
sardul3
Start the Jira-to-PR developer loop for a ticket.
SkillBFreshness: BClaude Code +9 - 0.524score
domain-glossary
sardul3
Maintain CONTEXT.md glossary and sparse ADRs.
SkillBFreshness: BClaude Code +9 - 0.524score
code-review
sardul3
Use after a feature, before merge or PR, or when the user asks for a review.
SkillBFreshness: BClaude Code +9 - 0.523score
systematic-debugging
sardul3
Use when hitting a bug, test failure, or unexpected behavior, before proposing a fix.
SkillBFreshness: BClaude Code +9 - 0.521score
dev-loop-review
sardul3
Staff review for /dev-loop.
SkillBFreshness: BClaude Code +9 - 0.521score
lavish-ui
sardul3
Use Lavish (npx -y lavish-axi) when config lavish.enabled is on for this repo (default auto = UI repos only) and the Jira story is visual.
SkillBFreshness: BClaude Code +9 - 0.521score
prompt-contract
sardul3
Fill the nine-slot prompt contract, then grill, then hand off to a clean Claude session.
SkillBFreshness: BClaude Code +9 - 0.519score
jira-progress
sardul3
Move a Jira issue across configured workflow columns and comment as /dev-loop progresses.
SkillBFreshness: BClaude Code +9 - 0.519score
pr-comment-fixer
sardul3
Fix GitHub PR review comments for a watched /dev-loop PR.
SkillBFreshness: BClaude Code +9 - 0.519score
repo-memory
sardul3
Load or rebuild the hashed repo index for /dev-loop.
SkillBFreshness: BClaude Code +9
Signals
- Stars
- 0
- Forks
- 0
- Open issues
- 0
- Last commit (this folder)
- Aug 30, 2026 (1 month ago)
From GitHub (fetched Oct 24, 2026).
Details
- License
- MIT
- Repository
- github.com/sardul3/dev-conductor
- First seen
- Oct 24, 2026
- Last checked
- Oct 24, 2026
- Source
- GitHubgithub-graphql
More from sardul3
- 0.524score
code-review
sardul3
Use after a feature, before merge or PR, or when the user asks for a review.
SkillBFreshness: BClaude Code +9 - 0.524score
session-handoff
sardul3
Compact this conversation into a handoff file for a fresh agent.
SkillBFreshness: BClaude Code +9 - 0.524score
domain-glossary
sardul3
Maintain CONTEXT.md glossary and sparse ADRs.
SkillBFreshness: BClaude Code +9 - 0.524score
design-tree-interview
sardul3
Relentless design-tree interview.
SkillBFreshness: BClaude Code +9 - 0.524score
dev-loop
sardul3
Start the Jira-to-PR developer loop for a ticket.
SkillBFreshness: BClaude Code +9 - 0.523score
systematic-debugging
sardul3
Use when hitting a bug, test failure, or unexpected behavior, before proposing a fix.
SkillBFreshness: BClaude Code +9
Cite this page
agent-memory skill. Agent Mart. https://agentmart.dev/skills/sardul3-agent-memory (last updated 2026-10-24)