maintaining-agent-guidance
Use when finishing or merging a substantial project change that may require durable updates to agent guidance.
Skill by jhonDoe15
Chalk Labs
Need help deploying maintaining-agent-guidance?
Chalk Labs sets up agents like this inside real teams: connected to your tools, tested on your data, with a human in the loop where it matters.
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/jhondoe15-maintaining-agent-guidance
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: 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/maintaining-agent-guidance
What's inside
- Skill name
maintaining-agent-guidance- Description
- Use when finishing or merging a substantial project change that may require durable updates to agent guidance.
- License
- MIT
- Instructions
- 105 lines in SKILL.md · view on GitHub
Files (4)
- SKILL.md
- component.json · evals/
- package-closure.json · evals/
- role.json · evals/
SKILL.md (first lines)
# Maintaining Agent Guidance Catch durable agent-facing knowledge at the final change boundary without turning every commit or pull request into a documentation pass. ## Enter once Enter when a substantial feature or project change has a stable outcome and is finishing or approaching merge. Require the effective implementation diff and validation evidence. Treat the completed change, not its Git operation, as the boundary. Mid-feature work, ordinary commits, PR creation, and routine changes do not qualify. Guidance edits from this run belong to the same change and do not retrigger the gate. ## Test significance Guidance needs inspection only when both are true: 1. Without the candidate knowledge, a future agent could make a materially worse project decision. 2. The knowledge is not cheaply and reliably derivable from current source or configuration. Likely candidates include: - a new or changed service scope, owner, or architectural boundary; - a non-obvious build, test, deployment, or environment procedure; - a compatibility, security, operational, or data-integrity invariant; - a recurring failure mode with an observable completion check; - a new instruction load path or conditional reference branch. Usually reject implementation summaries, changed-file lists, code-visible APIs, one-off debugging details, ticket status, release notes, and routine fixes or refactors. A fact already canonical in an ADR or maintainer document still passes when loaded guidance lacks the triggered route agents need to find it. When the test fails, return exactly: ```text Guidance: no change Reason: <one sentence> ``` Continue the finishing flow without loading or invoking `agents-file-writer`. ## Build the change-context packet When the test passes, collect a compact packet from the stable change: - feature or change outcome and why it matters; - originating requirements, issue, spec, or handoff references; - stable implementation diff or immutable commit range, plus available implementation and review handoffs; - validation evidence and affected service or project scope; - candidate durable knowledge, supporting evidence, and representative future task triggers; - unchanged and out-of-scope areas; - current Git and worktree state, including authorization limits. Reference durable artifacts and exact diff ranges instead of pasting broad repository context. The packet is complete when a worker can verify the candidate without reconstructing this session. ## Dispatch one isolated worker Dispatch exactly one isolated subagent. It may share the checkout, but not the main conversation context. Never run another writer concurrently in the same checkout. Give the worker the complete packet and this mandate: 1. Verify the candidate against repository authority; dispatch alone does not prove guidance must change. 2. If no durable gap exists, return the two-line no-change result and make no edits. 3. If a durable gap exists, invoke the required sub-skill `agents-file-writer` in **Improve** mode and follow its complete workflow. That skill owns file discovery, hierarchy, in-place edits, reference pointers, verification, and the artifact map. 4. Preserve the index and unrelated worktree changes. Leave staging, committing, pushing, publishing, PR mutation, and merging to the finishing flow unless separately authorized. 5. Return only the checkpoint below. Stop rather than guess at a collision or destructive-operation authorization boundary. The worker must improve durable guidance, not copy the feature summary into an agent file. If isolated subagents are unavailable, run only the significance test inline first. Load `agents-file-writer` inline only after it passes, then follow that skill in Improve mode. ## Resume the finishing flow Accept one of two checkpoints. For no change, use the exact result under **Test significance**. Changed or blocked: ```text Guidance: changed | blocked Changed paths: Guidance graph: Keep/Move/Delete: Validation: Untested gaps: Authorization blockers: ``` Verify the checkpoint against the checkout. When guidance changed, include those paths in the effective diff and rerun affected verification and review. Then resume the original finish or merge flow. This skill does not stage, commit, push, publish, mutate a PR, or merge.
From jhonDoe15/skills, MIT.
More from jhonDoe15/skills
- 0.570score
model-routing
jhonDoe15
Use before the first search or edit of work that needs more than two steps, and whenever bounded work might be handed to another model through `pi-dispatch`.
SkillAFreshness: AClaude Code +9 - 0.564score
incident-investigation
jhonDoe15
Evidence-led, read-only investigation of production incidents and hard-to-localize system or application bugs.
SkillAFreshness: AClaude Code +9 - 0.561score
skill-writing
jhonDoe15
Agent Skill outcome for creating or revising Agent Skills; excludes general agent-facing artifacts, fresh-context handoffs, installation, publication, and…
SkillAFreshness: AClaude Code +9 - 0.538score
readme-writing
jhonDoe15
Use when the user asks to write, rewrite, restructure, review, or correct a project's README.
SkillAFreshness: AClaude Code +9 - 0.529score
engineering-guidance
jhonDoe15
Engineering assessment of an existing technical approach or change for maintainability and standards; also used by declared design, implementation, and review…
SkillBFreshness: BClaude Code +9 - 0.526score
implementation-planning
jhonDoe15
Plan one bounded implementation ticket before code: produce a reviewed technical design and executable consumer-behavior TDD plan for another agent.
SkillBFreshness: BClaude Code +9 - 0.526score
architecture-review
jhonDoe15
Architecture review of existing code, a supplied technical design, or a PR change range.
SkillBFreshness: BClaude Code +9 - 0.524score
pr-carver
jhonDoe15
Use when assessing an existing Git branch or pull request and recommending one PR, parallel PRs, or an ordered stack from its prerequisites, migration…
SkillBFreshness: BClaude Code +9 - 0.524score
code-review
jhonDoe15
Reviews a complete Ticket outcome through fresh Domain and Engineering/Design lenses, then returns an evidence-backed read-only Review brief.
SkillBFreshness: BClaude Code +9 - 0.524score
skill-evaluation
jhonDoe15
Private evaluation dependency for Agent Skill behavior claims.
SkillBFreshness: BClaude Code +9 - 0.524score
implement
jhonDoe15
Use when a ready implementation blueprint and bounded scope need one isolated, TDD-driven patch with validation and a correction-ready implementation handoff.
SkillBFreshness: BClaude Code +9 - 0.523score
slice-plan
jhonDoe15
Private set-level decomposition consumed by Carve; not a user-facing planning outcome.
SkillBFreshness: BClaude Code +9
Signals
- Stars
- 0
- Forks
- 0
- Open issues
- 16
- Last commit (this folder)
- Sep 6, 2026 (1 month ago)
From GitHub (fetched Oct 24, 2026).
Details
- License
- MIT
- Repository
- github.com/jhonDoe15/skills
- First seen
- Oct 24, 2026
- Last checked
- Oct 24, 2026
- Source
- GitHubgithub-graphql
More from jhonDoe15
- 0.570score
model-routing
jhonDoe15
Use before the first search or edit of work that needs more than two steps, and whenever bounded work might be handed to another model through `pi-dispatch`.
SkillAFreshness: AClaude Code +9 - 0.564score
incident-investigation
jhonDoe15
Evidence-led, read-only investigation of production incidents and hard-to-localize system or application bugs.
SkillAFreshness: AClaude Code +9 - 0.561score
skill-writing
jhonDoe15
Agent Skill outcome for creating or revising Agent Skills; excludes general agent-facing artifacts, fresh-context handoffs, installation, publication, and…
SkillAFreshness: AClaude Code +9 - 0.538score
readme-writing
jhonDoe15
Use when the user asks to write, rewrite, restructure, review, or correct a project's README.
SkillAFreshness: AClaude Code +9 - 0.529score
engineering-guidance
jhonDoe15
Engineering assessment of an existing technical approach or change for maintainability and standards; also used by declared design, implementation, and review…
SkillBFreshness: BClaude Code +9 - 0.526score
architecture-review
jhonDoe15
Architecture review of existing code, a supplied technical design, or a PR change range.
SkillBFreshness: BClaude Code +9
Done for you
Rather not set it up yourself?
Chalk Labs scopes, builds and runs agents for your team: the task audit, the eval harness, the integrations and the weekly ops report. Fixed scope, priced before we start.
See how we build AI agents →Read the case studies
Chalk Labs is a two-founder AI and growth studio. We build AI agents, CRMs and MVPs for Web3 and AI companies, and have shipped 50+ product and token launches.
- launches shipped
- 50+
- launches shipped
- media placements
- 500+
- media placements
- founders on every account
- 2
- founders on every account
Press we've placed: Forbes · CoinDesk · Decrypt · Cointelegraph · The Block
Rahil Jain, Co-founder, Chalk Labs. Ex-Goldman Sachs engineer. You talk to the person who builds it.
Get maintaining-agent-guidance running in your business
Tell us what you want to automate. A founder replies within one working day with a plan, not a pitch deck.
Cite this page
maintaining-agent-guidance skill. Agent Mart. https://agentmart.dev/skills/jhondoe15-maintaining-agent-guidance (last updated 2026-10-24)